Optimized irrigation pressure control system and method for rigid choledochoscope based on visual field blurring

Abstract

This invention discloses a rigid cholangioscopy irrigation pressure control system and method based on visual field ambiguity optimization, belonging to the field of medical device and process control technology. It constructs a multi-dimensional perception feature vector by simultaneously acquiring instantaneous pressure within the bile duct and the original video stream; it quantifies red staining, particles, bubbles, and tissue adhesion in real time, calculating a comprehensive evaluation value of visual obstruction; based on this, it retrieves a strategy expert database to generate control parameters, and drives a dual-pump system to output biomimetic pulsed fluid of a specific frequency through adaptive pulse width modulation, efficiently stripping and removing impurities; by predicting future pressure trends, it triggers emergency pressure release in advance when a safety threshold is reached. This invention achieves a deep integration of accurate identification of ambiguity causes, efficient biomimetic pulsed debridement, and active overpressure defense, significantly improving visual field recovery efficiency and surgical continuity while maintaining a safe intra-bile duct environment.

Description

Technical Field

[0001] This invention belongs to the field of medical device and process control technology, and specifically relates to a rigid cholangioscopy perfusion pressure regulation system and method based on visual field ambiguity optimization. Background Technology

[0002] Rigid choledochoscopes are core medical instruments in hepatobiliary surgery for bile duct exploration, stone removal, and polyp resection. Maintaining a clear endoscopic view and a stable bile duct environment during the procedure is crucial for surgical success and preventing bile duct injury. The irrigation system continuously injects water to dilate the bile duct and flush away dislodged impurities; its hydrodynamic properties and control mechanisms directly determine the continuity of the surgery and the final clinical outcome.

[0003] Existing cholangioscopic irrigation systems are primarily used in high-frequency clinical scenarios such as common bile duct incision and stone removal, electrohydraulic lithotripsy, or laser lithotripsy. Traditional techniques often employ constant pressure irrigation or gravity-based dripping, with some high-end devices equipped with manually adjustable roller pumps. In practical applications, when the surgeon activates the lithotripsy equipment, a large amount of stone powder and air bubbles are generated within the bile duct, or capillary bleeding occurs due to instrument friction, causing the endoscopic field of view to be rapidly obscured and blurred. At this point, the surgeon must interrupt the core surgical procedure and manually increase the irrigation pump's speed or pressure to forcibly flush the field of view. Technical analysis shows that existing equipment only passively measures a single physical pressure value as a feedback indicator, completely lacking the ability to perceive the "visual quality" of the endoscopic image. The irrigation system cannot distinguish whether the blurred field of view is caused by high-density blood, reflective air bubbles, or physical tissue obstruction, and can only mechanically use a single and constant water flow for blind flushing, resulting in extremely low hydrodynamic replacement efficiency.

[0004] This reveals the following problems with the existing technology: First, the feedback is delayed, relying on manual adjustment of the equipment after visually detecting blurriness, which can easily disrupt the surgical rhythm; second, blind flushing makes it difficult for the system to match the optimal flow field mode for different causes of blurriness, resulting in low efficiency in removing impurities; third, there are significant safety risks, as blindly and continuously increasing pressure in pursuit of a clear field of vision can easily exceed the safe physiological threshold of 30cm H2O in the bile duct, causing bile reflux into the blood and even leading to fatal biliary sepsis. Summary of the Invention

[0005] (a) Technical problems to be solved To address the problems in related technologies, this invention provides a rigid cholangioscope perfusion pressure control system and method based on visual field blur optimization, in order to overcome the aforementioned technical problems existing in the prior art.

[0006] (II) Technical Solution To solve the above-mentioned technical problems, the present invention is achieved through the following technical solution: S1. Simultaneously acquire the original pressure signal and original video stream inside the bile duct in real time, perform multi-scale feature preprocessing on the original video stream to generate an image sequence to be analyzed containing spatial gradient and frequency domain distribution features, and extract multi-dimensional sensing feature vectors at the same time. S2. Input the image sequence to be analyzed and the multidimensional perceptual feature vector into the visual blurring cause decoupling model, combine it with the current pressure signal, quantify the red staining degree, particle suspension degree, bubble interference degree and tissue adhesion degree in real time, and calculate the comprehensive evaluation value of visual impairment accordingly. S3. Based on the comprehensive evaluation value of visual obstruction, the quantitative results of each ambiguity cause, and the current pressure value, retrieve the perfusion strategy expert database and generate a preliminary perfusion control parameter set including flushing intensity, fluid replacement rate, and target pressure range. S4. Based on the preliminary infusion control parameter set, the coordinated speed curve of the inlet pump and the outlet pump is generated by adaptive pulse width modulation, and the biomimetic pulse fluid is output to perform impurity stripping and discharge, and the feedback pressure and real-time image after execution are obtained. S5. Use a long short-term memory network to make rolling predictions of future pressure and calculate the distance between image feature vectors before and after regulation to obtain the visual improvement rate; when the predicted pressure reaches the safety threshold or the visual improvement rate fails to meet the standard for a continuous period of time, trigger the second correction of the regulation parameters and the emergency pressure relief procedure. This invention, by synchronizing visual and pressure signals in real time, not only achieves deep decoupling and precise quantification of the causes of blurred vision, but also, by introducing a biomimetic pulse flushing and LSTM pressure prediction mechanism, changes the passive drawbacks of traditional blind flushing and post-operative alarms, enabling diagnosis and treatment as well as pre-operative intervention. While efficiently removing impurities and rapidly restoring the surgical field of vision, it strictly limits the intra-biliary pressure within a safe physiological threshold, reducing the risk of complications such as biliary pancreatitis and sepsis, and improving surgical continuity and safety.

[0007] Preferably, step S1 includes the following steps: S11. Using a piezoresistive miniature pressure sensor integrated into the tip of the rigid cholangioscope sheath, the instantaneous pressure in the bile duct is collected, and the baseline drift is removed by median filtering to obtain a smooth pressure time series. S12. Obtain the original video stream through the photoelectric conversion module, calculate the global gradient variance of each frame image using the Laplacian operator, and use it as the initial sharpness reference value to complete the initial image quality judgment. S13. After the initial quality judgment, the image is grayscaled and histogram equalized to enhance the edges of debris and bubbles. The high-frequency component energy ratio is extracted by fast Fourier transform, and the image is subjected to local contrast analysis to mark suspected blur candidate areas. S14. Align the global gradient variance, high-frequency energy ratio, suspected blurred candidate region labeling information with the smoothed pressure signal along the time axis to construct a multi-dimensional perceptual feature vector and simultaneously form an image sequence to be analyzed. By aligning the underlying timestamps of the piezoresistive micro-sensor with those of the high-definition image module, a high-fidelity multidimensional sensing feature vector was constructed. Median filtering and the Laplacian operator were used to effectively eliminate instrument collision noise and baseline drift, ensuring the purity of the underlying data. Multi-scale feature preprocessing transformed a single visual image into structured data containing spatial gradients and frequency domain energy, providing a data foundation for subsequent fuzzy decoupling and pressure control.

[0008] Preferably, the step of marking suspected fuzzy candidate regions in S13 includes: S131. Set a pixel sliding window, traverse the images in the original video stream based on the pixel sliding window, calculate the local contrast factor, and mark the areas with contrast below the threshold as suspected blur candidate areas. S132. Set a global average density threshold; perform Sobel edge detection on suspected blurry candidate areas, count the edge pixel density, and if the edge density of a certain area is lower than the global average density threshold, it is determined that there is substantial visual occlusion in the area, and the determination information is incorporated into the multidimensional perception feature vector. By using dual verification of local contrast factor and Sobel edge density, the system accurately locates the substantial visual occlusion area; it effectively eliminates interference from local reflections or normal tissue physiological textures, avoids resource waste and misjudgment caused by global calculation, and greatly improves the noise resistance of fuzzy area localization and the real-time performance of system operation.

[0009] Preferably, step S2 includes the following steps: S21. Input the image sequence to be analyzed into the visual blurring cause decoupling model; obtain the red occlusion degree through red channel pixel ratio analysis; S22. Use morphological closing operations to identify circular or elliptical bright patches in the image, calculate the number of patches and the movement vector, and obtain the bubble interference degree. S23. Analyze the high-frequency noise distribution pattern of the image, identify suspended particles with a diameter smaller than a preset threshold and calculate their coverage area ratio to obtain the particle suspension degree. S24. Combining the fluctuation characteristics of the pressure time series, when the pressure suddenly increases and a large area of ​​low texture appears on the screen, determine the tissue adhesion and calculate the tissue adhesion degree. S25. The red staining opacity, bubble interference, particle suspension and tissue adhesion are weighted and fused to obtain a comprehensive evaluation value of visual obstruction. By constructing a decoupling model for the causes of visual blurring, the general problem of unclear vision is refined into four specific physical causes: red staining, air bubbles, particles, and tissue adhesion. This breaks through the limitations of traditional perfusion systems that rely solely on a single pressure threshold for regulation. By extracting features and combining them with pressure mutation signals, precise three-dimensional quantification of the complex bile duct environment is achieved, providing a scientific and objective decision-making basis for subsequent matching of differentiated hydrodynamic flushing strategies.

[0010] Preferably, the calculation of tissue adhesion in S24 includes: S241. Extract the mutual information entropy of ten consecutive frames of images. When the slope of the mutual information entropy decreases exceeds the threshold and the pressure signal shows increased resistance, start tissue adhesion-assisted recognition. S242. Calculate the motion vector in the central region of the image using the optical flow algorithm. If the vector length approaches zero and the pixel grayscale distribution is highly consistent, it is confirmed that the lens is completely blocked by the tube wall or a large piece of tissue, and the tissue adhesion is assigned the highest level. If the condition is not met, it is linearly mapped to a value between 0 and 1 based on the values ​​of the motion vector length and gradient variance. By fusing the motion vectors of the optical flow algorithm with mutual information entropy and pressure mutation features, the system accurately distinguishes between two dangerous scenarios that appear similar but have drastically different physical states: lens contamination and tissue adhesion (blind end). This avoids the risk of blindly applying pressure and bile duct perforation caused by misjudgment.

[0011] Preferably, step S3 includes the following steps: S31. Calculate the minimum flushing volume required to replace the original fluid in the bile duct based on the red staining degree, and use it as the basic flow compensation value of the inlet pump. S32. Based on the particle suspension degree and bubble interference degree, match the corresponding flow field disturbance mode from the strategy expert library, including laminar displacement mode, turbulent scouring mode and intermittent negative pressure attraction mode. S33. Set the target pressure range and calculate the pressure adjustment ratio coefficient based on the deviation between the current pressure and the target range; S34. Couple the basic flow compensation value, the matched flow field disturbance mode and the pressure regulation proportional coefficient to generate a preliminary set of control parameters including pump head speed, start-stop cycle and the difference in flow between internal and external pumps. S35. Introduce an environmental compliance factor that reflects the degree of bile duct dilation to dynamically correct the flow gain in the initial control parameter set; By connecting the sensing module and the execution module, a precise mapping from visual evaluation values ​​to fluid dynamic parameters is achieved. Relying on the strategy expert database, the system can automatically match the optimal flow field mode such as laminar flow, turbulent flow or pulse for different ambiguity causes. At the same time, the introduction of environmental compliance factors to dynamically correct the flow rate ensures that the perfusion strategy can maximize flushing efficiency and adapt to the differences in bile duct elasticity between individual patients, thus achieving personalized and flexible parameter output.

[0012] Preferably, step S4 includes the following steps: S41. The preliminary control parameter set is converted into the PWM duty cycle signal of the water inlet pump controller to control the water inlet pump to generate sawtooth wave pressure pulses and use fluid impact force to peel off the adhering impurities. S42. Set the water flow rate multiple; synchronously adjust the suction power of the outlet pump to keep the outlet flow rate at a multiple of the inlet flow rate, so as to maintain the bile duct lumen filling stably; S43. Monitor the pressure change rate in the bile duct in real time after execution, and adjust the duty cycle of the pulse waveform through the feedback loop; S44. Set the red dye masking threshold; when the red dye masking is detected to drop below the red dye masking threshold, switch the pulse rinsing to the steady laminar flow mode. The water hammer effect and shear force generated by the biomimetic sawtooth wave pulse are far more efficient at removing stubborn sludge and blood clots from the bile duct than traditional constant pressure continuous flushing. At the same time, the proportional following suction mechanism of the water pump maintains the dynamic balance of the volume of the bile duct cavity while ensuring high-flow and powerful flushing, thus avoiding bile duct wall collapse or spasm caused by drastic pressure fluctuations.

[0013] Preferably, the feedback loop adjustment in S43 includes: S431. Calculate the ratio of the current pressure change rate to the safe pressure change threshold. If the ratio is greater than 1, reduce the amplitude of the water inlet pulse proportionally. S432. Calculate the slope of the improvement in visual clarity over the past three pulse cycles. If the slope is less than the optimization target value, gradually increase the pulse frequency within the safety pressure limit. Through a dual closed-loop feedback mechanism of pressure change rate and clarity enhancement slope, the pulse parameters are self-calibrated, ensuring that the rinsing force always hovers at the critical optimal solution between cleaning efficiency and physiological safety pressure, preventing insufficient cleaning or excessive stimulation.

[0014] Preferably, step S5 includes the following steps: S51. Input real-time pressure data into the long short-term memory network prediction model to calculate the pressure evolution trend line at future moments; S52. If the trend line touches the safety red line, immediately output a forced pressure reduction command, shut down the inlet pump and start the outlet pump at full speed. S53. Set a step size deviation threshold, calculate the distance between the image feature vectors before and after adjustment as the visual improvement rate. If the visual improvement rate is lower than the step size deviation threshold within five consecutive adjustment cycles, issue a manual intervention prompt. S54. Store the successful parameter combination of this adjustment into the local memory and update the weight parameters of the strategy expert database online. By introducing a dual monitoring mechanism of LSTM timing prediction and visual improvement rate, the system has achieved a leap from passive defense of pressure relief after overpressure to active defense of pre-overpressure cutoff. Through trend prediction and forced pressure reduction commands, a safety baseline has been established. At the same time, the local storage of successful parameters and online updates of the expert database have given the system the ability to continuously evolve, making it more and more accurate with the increase of usage.

[0015] Preferably, a rigid cholangioscopy perfusion pressure control system based on field-of-view optimization is used to implement the rigid cholangioscopy perfusion pressure control method based on field-of-view optimization. The system includes: The multidimensional data acquisition module is used to acquire and preprocess intraductal pressure signals and video image streams in real time, and output pressure time series, image sequence to be analyzed and multidimensional sensing feature vectors. The visual feature analysis module receives the image sequence to be analyzed and the stress time series, decouples the causes of visual blurring and calculates the comprehensive evaluation value of visual impairment; The control decision center module generates the optimal control parameter set based on the comprehensive evaluation value of visual obstruction, the classification of ambiguity causes, and the current pressure. The adaptive execution drive module drives the inlet and outlet water pumps to generate a coordinated biomimetic pulse flow field based on the optimal control parameter set and feeds back the data after execution. The safety monitoring and feedback module uses the visual improvement rate calculated from the stress prediction and the multi-dimensional perception feature vector to perform closed-loop safety monitoring and strategy updates.

[0016] (III) Beneficial Effects The present invention has the following beneficial effects: This invention achieves decoupling and precise perception of the causes of ambiguity. Through multi-scale feature preprocessing and input visual ambiguity cause decoupling model, the system can not only quantify the degree of ambiguity in the field of vision in real time, but also accurately distinguish four specific physical causes: red staining, particles, bubbles, and tissue adhesion. This diagnostic and treatment-based perception mechanism breaks through the limitations of blind washing in traditional equipment, enabling the perfusion system to automatically match the optimal fluid dynamics strategy according to the physical characteristics of different impurities, thereby improving the field of vision recovery efficiency in complex scenarios.

[0017] This invention constructs an adaptive biomimetic pulse flushing and dual-pump collaborative execution mechanism; utilizing the water hammer effect and extremely high fluid shear force generated by sawtooth wave pulses, the system can efficiently remove stubborn biliary sludge and blood clots adhering to the wall; at the same time, through the proportional follow-up suction of the water pump and the dynamic correction of environmental compliance factors, the system can still maintain the dynamic balance of the volume of the bile duct lumen when performing high-flow and powerful flushing; this not only avoids the collapse of the bile duct wall due to drastic pressure fluctuations, but also ensures the continuous stability of the surgical operating space.

[0018] This invention constructs a proactive safety defense system based on time series prediction. By introducing an LSTM network to predict future intrabiliary pressure, the system achieves a transition from pressure relief after overpressure to pre-overpressure cutoff. Once the pressure is predicted to reach the safety threshold, a forced dual-pump pressure relief command is immediately triggered, avoiding the risk of bile duct rupture and biliary infection. In addition, the closed-loop evaluation and local storage mechanism of the control parameters endows the expert database with the ability to be continuously updated, making the system increasingly intelligent and accurate in long-term clinical applications.

[0019] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0020] To more clearly illustrate the technical solutions of the embodiments of the invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the invention. For those skilled in the art, the drawings can be obtained from these drawings without creative effort.

[0021] Figure 1 This is a flowchart illustrating the method for controlling perfusion pressure in a rigid cholangioscope based on visual field ambiguity optimization according to the present invention. Figure 2 This is a schematic diagram of the process for calculating the comprehensive evaluation value of visual impairment in this invention; Figure 3 This is a schematic diagram of the process for obtaining the preliminary perfusion control parameter set in this invention; Figure 4 This is a schematic diagram of the impurity stripping and discharge process in this invention; Figure 5 This is a schematic diagram of the module of the rigid cholangioscopy perfusion pressure control system based on field of view ambiguity optimization of the present invention. Detailed Implementation

[0022] The technical solutions of the embodiments of the invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the invention, and not all embodiments. Based on the embodiments of the invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the invention.

[0023] Please see Figure 1 This invention discloses a method for regulating perfusion pressure in a rigid cholangioscope based on visual field ambiguity optimization, comprising the following steps: S1. Simultaneously acquire the original pressure signal and original video stream inside the bile duct in real time, and perform multi-scale feature preprocessing on the original video stream to generate a multi-dimensional sensing feature vector containing spatial gradient and frequency domain distribution features. S1 includes the following steps: S11. Using a piezoresistive miniature pressure sensor integrated into the tip of the rigid cholangioscope sheath, instantaneous pressure in the bile duct is collected at a sampling frequency of not less than 1000Hz. After removing baseline drift by median filtering, a smooth pressure time series is obtained. Specifically, S11 involves embedding a MEMS micro pressure-sensitive chip at the front end of the 3mm outer diameter sheath of a rigid choledochoscope. This chip has a measurement range of 0-100cmH2O. The system performs analog-to-digital conversion (ADC) at a frequency of 1024Hz. The acquired raw voltage signal first passes through a fifth-order median filter to remove high-frequency pulse noise caused by surgical instrument collisions or doctor's hand tremors. Subsequently, an atmospheric pressure offset is removed through a baseline compensation algorithm to ensure that the zero-point reference of the pressure data is always at the reference plane of the biliary environment. S12. Obtain the 1080P / 60fps raw video stream through the photoelectric conversion module, and use the Laplacian operator to calculate the global gradient variance of each frame image as the initial sharpness reference value to complete the initial image quality judgment. S12 specifically involves: the image processor receiving the raw RAW data transmitted back from the CMOS sensor via the 3G-SDI interface and converting it into an RGB space image; the system performing convolution operations on each frame of the image using a 3x3 Laplacian template and calculating the gradient variance of all pixels in the image; when the variance value is lower than the threshold of 500, it is preliminarily determined that the field of view has entered a blurred state. S13. After the initial quality judgment, the image is grayscaled and histogram equalized to enhance the edges of debris and bubbles. The high-frequency component energy ratio is extracted by fast Fourier transform, and the image is subjected to local contrast analysis to mark suspected blur candidate areas. The step of marking suspected fuzzy candidate regions in S13 includes: S131. Set a 32×32 pixel sliding window, traverse the images in the original video stream based on the pixel sliding window, calculate the local contrast factor, and mark the areas with contrast below the threshold as suspected blur candidate areas. S132. Set a global average density threshold; perform Sobel edge detection on suspected blurry candidate areas, and count the edge pixel density. If the edge density of a certain area is lower than 30% of the global average density threshold, it is determined that there is substantial visual occlusion in the area, and this determination information is incorporated into the multidimensional perception feature vector. In specific implementation, S13 is as follows: The system divides the image into 1200 local windows and uses CLAHE (Contrast Limiting Adaptive Histogram Equalization) technology to enhance local details; by observing the spectrum through FFT transformation, if the energy attenuation in the high-frequency region exceeds 6dB, it indicates that the image details are severely lost; at this time, step S132 further confirms this by using the Sobel operator, and if the edge pixel ratio is less than 5%, then the region is locked as the key washing target area; S14. Align the global gradient variance, high-frequency energy ratio, suspected blurred candidate region labeling information with the smoothed pressure signal along the time axis to construct a multi-dimensional perceptual feature vector and simultaneously form an image sequence to be analyzed. S14 specifically involves: the system establishing a unified timestamp management queue, interpolating and aligning the image features of 60 frames per second with 1024 pressure sampling points per second to form a feature vector containing [time, pressure, gradient variance, high-frequency energy, local contrast]. The multidimensional perceptual feature vector and the image sequence to be analyzed together serve as the input benchmark for the fuzziness cause analysis in S2; By simultaneously executing high-frequency pressure sampling and multi-scale image feature preprocessing, a time-strictly aligned multidimensional perception feature vector was constructed. This solved the problems of pressure feedback lag and temporal misalignment between visual and mechanical states in traditional perfusion systems from the data source, providing a complete data foundation for the precise decoupling of subsequent ambiguity causes. This enabled the system to perceive abnormal visual states with both global evaluation capabilities and precise local positioning capabilities. S2. Input the image sequence to be analyzed and the multidimensional perceptual feature vector into the visual blurring cause decoupling model, combine it with the current pressure signal, quantify the red staining degree, particle suspension degree, bubble interference degree and tissue adhesion degree in real time, and calculate the comprehensive evaluation value of visual impairment accordingly. Please see Figure 2 S2 includes the following steps: S21. Input the image sequence to be analyzed into the visual blur cause decoupling model; obtain the red shading degree through red channel pixel ratio analysis; the visual blur cause decoupling model is a two-stream convolutional neural network, including a chroma stream branch and a texture stream branch; Specifically, S21 involves: real-time monitoring of the H (hue) and S (saturation) channels in the HSV color space via the chroma stream branch; when the proportion of pixels in the red region (H values ​​in the range of 0-10 and 160-180) exceeds 15% of the entire image, calculating its average saturation to obtain the red shading degree. D red ;like D red A value exceeding 0.6 indicates significant active bleeding. Furthermore, the construction and training of the visual blur cause decoupling model includes the following steps: 211. Obtain image sequence data and corresponding multidimensional perceptual feature vectors (including pressure, gradient variance, high-frequency energy ratio, etc.) from several historical rigid cholecystoscopic surgeries as input samples; clinical experts quantify the causes of visual blur in the image sequences, and score the red staining occlusion, particle suspension, bubble interference, and tissue adhesion with continuous values ​​between 0 and 1 to construct a historical visual blur feature dataset and a historical blur cause label dataset. 212. Based on the historical visual blur feature dataset and the historical blur cause label dataset, construct a visual blur cause decoupling model; Furthermore, the construction and training of the visual blur cause decoupling model includes the following steps: 2121. Construct an initial visual blur cause decoupling model and set a first training data ratio; divide the historical visual blur feature dataset and the historical blur cause label dataset according to the first training data ratio to obtain a first training dataset and a first test dataset. S2122. Set a first training error threshold; input the first training dataset into the initial visual blur cause decoupling model for training; during the training process, if the training error is less than the first training error threshold, stop training and obtain the trained visual blur cause decoupling model; otherwise, continue training until the training error is less than the first training error threshold. S2123. Set a first test error threshold; input the first test dataset into the trained visual blur cause decoupling model for testing; after the test is completed, obtain the first test error data; if the first test error data is less than or equal to the first test error threshold data, use the trained visual blur cause decoupling model as the final visual blur cause decoupling model; otherwise, return to S2122 to continue training the trained visual blur cause decoupling model and repeat S2123 until the first test error data is less than or equal to the first test error threshold. The first training data ratio, the first training error threshold, and the first test error threshold can be dynamically adjusted according to the model architecture, data volume, and actual needs during specific implementation. The training error and test error are both obtained by using historical fuzzy cause label data as the reference label and solving the difference between the model output data and the reference label data through cosine similarity or Euclidean distance calculation. Furthermore, the architecture of the initial visual blur cause decoupling model includes an input layer, a chroma flow branch, a texture flow branch, a feature fusion layer, and an output layer; Specifically, the input layer receives a sequence of images to be analyzed with a size of 224×224×3 and a multidimensional perceptual feature vector of length 5; the chroma stream branch contains three consecutive convolutional modules (each module contains a 3×3 convolutional layer, a batch normalization layer BatchNorm, and a ReLU activation function) and one global average pooling layer, focusing on extracting the red channel saturation and hue shift features of the image in the HSV color space; the texture stream branch contains four dilated convolutions. The Convolution layer consists of a convolutional module with an inflation rate of 2 and a max-pooling layer. Its input is superimposed with Laplacian gradient features and high-frequency energy ratio to expand the receptive field and capture the bright reflection closed contour features of small suspended particles and bubbles. The feature fusion layer uses the Concat operation to concatenate the 128-dimensional feature vector output from the chroma flow branch, the 256-dimensional feature vector output from the texture flow branch, and the smooth pressure signal features at the current time step. The feature dimensionality reduction and fusion are then performed through a fully connected layer with 256 neurons. The output layer contains 4 neurons, and the activation function is the Sigmoid function, which corresponds to four quantized values ​​between 0 and 1 for output red staining occlusion, particle suspension, bubble interference, and tissue adhesion. S22. Use morphological closing operations to identify circular or elliptical bright patches in the image, calculate the number of patches and the movement vector, and obtain the bubble interference degree. Specifically, S22 involves the following: bubbles exhibit typical high-brightness reflection characteristics under cold light illumination; the texture flow branch in the model extracts high-frequency feature blocks with brightness higher than 240 through shallow convolutional kernels, and its deep network implicitly performs feature connections similar to morphological closing operations; combined with the input of consecutive frames, the network extracts closed contour features with a major-minor axis ratio between 0.8 and 1.2 and a significant upward displacement vector, and finally the output layer calculates and outputs the bubble interference degree. D bubble ; S23. Analyze the high-frequency noise distribution pattern of the image, identify suspended particles with a diameter smaller than a preset threshold and calculate their coverage area ratio to obtain the particle suspension degree. Specifically, S23 is as follows: suspended particles (sludge, gravel) appear as small, randomly moving grayscale spots in the image; the texture flow branch effectively captures small high-frequency noise features with a diameter between 5 and 20 pixels through dilated convolution, and combines the high-frequency energy ratio parameter in the input multidimensional perceptual feature vector to dynamically regress the particle suspension degree in the current field of view in the model output layer. S24. Combining the fluctuation characteristics of the pressure time series, when the pressure suddenly increases and a large area of ​​low texture appears on the screen, determine the tissue adhesion and calculate the tissue adhesion degree. The calculation of tissue adhesion in S24 includes: S241. Extract the mutual information entropy of ten consecutive frames of images. When the slope of the mutual information entropy decreases exceeds the threshold and the pressure signal shows increased resistance, start tissue adhesion-assisted recognition. S242. Calculate the motion vector in the center region of the image using the optical flow algorithm. If the vector length approaches zero and the pixel grayscale distribution is highly consistent, it is confirmed that the lens is completely blocked by the tube wall or a large piece of tissue. Assign the tissue adhesion to the highest level and determine the tissue adhesion accordingly. If the condition is not met, linearly map it to a value between 0 and 1 based on the values ​​of the motion vector length and gradient variance. Specifically, S24 involves the following: When the doctor advances the endoscope too quickly or there is bile duct spasm, the lens often becomes stuck to the mucosa. At this time, the image input to the model appears uniformly flesh-colored with extremely weak texture features, and the pressure features received by the fusion layer show a surge from 20 cmH2O to over 40 cmH2O. The model combines the extremely low mutual information entropy (below 1.5 bits) extracted from preprocessing with the optical flow motion vector approaching zero, and through the nonlinear mapping of the feature fusion layer, determines that the current state is tissue adhesion. The output layer directly outputs the tissue adhesion degree. D tissue It is the highest level; S25. The red staining opacity, bubble interference, particle suspension and tissue adhesion are weighted and fused, with the red staining opacity having the highest weight, to obtain the comprehensive evaluation value of visual obstruction. Specifically, S25 refers to: comprehensive evaluation value. V =0.4× D red +0.2× D dust +0.15× D bubble +0.25× D tissueThe evaluation system fully considers the degree of interference of different ambiguity causes on clinical operation. The comprehensive evaluation value of visual impairment and the quantitative results of each ambiguity cause are output to S3 together. The weight coefficients in the comprehensive evaluation value calculation formula are obtained based on the data in the historical clinical surgical database to obtain the influence of red staining occlusion, bubble interference, particle suspension and tissue adhesion on the comprehensive evaluation value. By using a decoupling model for the causes of visual blurring to decouple chromatic and textural features in parallel, and combined with post-processing logic that uses pressure signals for auxiliary determination, the system achieves a four-dimensional, refined quantification of the causes of blurred vision in cholangioscopy (red staining obscuration, bubble interference, particle suspension, and tissue adhesion). This breaks through the limitations of traditional methods that can only determine whether the field of vision is blurred, enabling the system to accurately identify why the vision is blurred, and providing a clear clinical direction for the precise matching of subsequent differentiated control strategies. S3. Based on the comprehensive evaluation value of visual obstruction, the quantitative results of each ambiguity cause, and the current pressure value, retrieve the perfusion strategy expert database and generate a preliminary perfusion control parameter set including flushing intensity, fluid replacement rate, and target pressure range. Please see Figure 3 S3 includes the following steps: S31. Calculate the minimum flushing volume required to replace the original fluid in the bile duct based on the red staining degree, and use it as the basic flow compensation value of the inlet pump. Specifically, S31 involves setting the average volume of the common bile duct to 15 ml; if... D red The value is 0.5. The system calculates that it needs to complete the replacement of 3 times the volume (45ml) of liquid within 5 seconds, so the basic flow compensation of the inlet pump is set to 540ml / min. S32. Based on the particle suspension degree and bubble disturbance degree, match the corresponding flow field disturbance mode from the strategy expert database, including laminar displacement mode, turbulent scouring mode, and intermittent negative pressure attraction mode; specifically: S321, Laminar Flow Displacement Mode: Used for slight red staining, maintaining stable inflow and outflow; S322, Turbulent scouring mode: used for suspending a large amount of gravel, generating swirling flow by changing the inlet angle; S323, Intermittent negative pressure suction mode: Used for bubble accumulation, the bubbles are expelled by short suction. Specifically, S32 is as follows: the expert database stores parameter combinations of thousands of typical clinical scenarios; when the particle suspension degree exceeds 0.4, the system automatically calls the S322 mode, and through the asymmetric pump head drive algorithm, local turbulence is formed in the bile duct, so that the tiny stones deposited on the lower wall of the bile duct are suspended and discharged with the fluid. S33. Set 30cm H2O as the absolute safety upper limit, set the target pressure range as 15-25cm H2O, and calculate the pressure adjustment ratio coefficient based on the deviation between the current pressure and the target range; S34. Couple the basic flow compensation value, the matched flow field disturbance mode and the pressure regulation proportional coefficient to generate a preliminary set of control parameters including pump head speed, start-stop cycle and the difference in flow between internal and external pumps. In specific implementation, S34 is as follows: The system adopts a fuzzy logic controller and uses the outputs of S31 to S33 as input variables; for example, in the scenario of red dye superimposed high pressure, the controller will prioritize the strategy of large flow rate combined with equal amount of suction, that is, the speed of the inlet pump is adjusted to 3000 rpm and the speed of the outlet pump is adjusted to 3100 rpm, and the pressure rise caused by the high flow rate is offset by the slight negative pressure. S35. Introduce an environmental compliance factor that reflects the degree of bile duct dilation to dynamically correct the flow gain in the initial control parameter set; In specific implementation, S35 specifically involves: assessing bile duct compliance by measuring the diameter change of the bile duct contour in the image; if the bile duct does not expand significantly after increasing the flow rate, it indicates that the bile duct wall has poor elasticity or is already in a high tension state, at which point the flow rate gain coefficient is automatically reduced by 20% to prevent damage. By deeply coupling the fuzzy causal quantification results with the perfusion strategy expert database, different flow field disturbance modes such as laminar flow displacement, turbulent scouring, or intermittent negative pressure suction are automatically matched according to different causal combinations, and a preliminary set of control parameters is generated by combining pressure safety constraints. At the same time, environmental compliance factors are introduced for dynamic correction, so that the control parameters can adapt to the differences in the physiological state of the bile ducts of different patients, realizing the leap from uniform and extensive flushing to disease-specific treatment and dynamic control. S4. Based on the preliminary infusion control parameter set, the coordinated speed curve of the inlet pump and the outlet pump is generated by adaptive pulse width modulation, and the biomimetic pulse fluid is output to perform impurity stripping and discharge, and the feedback pressure and real-time image after execution are obtained. Please see Figure 4 S4 includes the following steps: S41. The preliminary control parameter set is converted into the PWM duty cycle signal of the water inlet pump controller, and the water inlet pump is controlled to generate sawtooth wave pressure pulses with a frequency of 2-5Hz, using fluid impact force to peel off the adhering impurities. Specifically, S41 refers to the shear force generated by a 2-5Hz pulse flow compared to a constant flow rate; the system controls the inlet pump to accelerate from 0 to 80% of the maximum flow rate within 200ms, maintain it for 50ms, and then quickly drop back; this hammering effect can effectively loosen the sludge adhering to the mucosa. S42. Set the water flow rate multiple to 90%-110%; synchronously adjust the suction power of the outlet pump to keep the outlet flow rate at a multiple of the inlet flow rate, thus maintaining stable filling of the bile duct. Specifically, S42 is as follows: the water pump adopts a follow-up control algorithm; when the water inlet pulse is at its peak, the water pump starts compensation suction after a 10ms delay to ensure that the instantaneous peak value of the pressure inside the bile duct does not exceed 35cm H2O and the average pressure is maintained at around 20cm H2O. S43. Monitor the pressure change rate in the bile duct in real time after execution, and adjust the duty cycle of the pulse waveform through the feedback loop; The feedback loop adjustment in S43 includes: S431. Calculate the ratio of the current pressure change rate to the safe pressure change threshold. If the ratio is greater than 1, reduce the amplitude of the water inlet pulse proportionally. S432. Calculate the slope of the improvement in visual clarity over the past three pulse cycles. If the slope is less than the optimization target value, gradually increase the pulse frequency within the safety pressure limit. The slope of the improvement in visual clarity is calculated by the gradient variance evolution extracted in S1; Specifically, S432 is as follows: if the image gradient variance only increases by 10% after three consecutive pulses, failing to reach the expected 30% target, the system determines that the current flushing intensity is insufficient; under the premise that the bile duct baseline pressure is detected to be lower than 25cm H2O, the pulse frequency is increased from 2Hz to 4Hz to enhance the debridement effect by increasing the number of impacts per unit time. S44. Set the red staining masking threshold; when the red staining masking level is detected to drop below the red staining masking threshold, automatically switch the pulse flushing to a stable laminar flow mode to reduce mucosal irritation. The biomimetic sawtooth wave pulse fluid generated by adaptive pulse width modulation peels off adhering impurities with high shear force generated by the hammering effect. At the same time, the water pump adopts a follow-up control algorithm to maintain the bile duct filling steady state. This step transforms the decision parameters of S3 into precise waveform control of physical execution. Under the closed-loop adjustment of slope feedback to improve visual clarity, a dynamic balance between debridement force and pressure safety is achieved, avoiding the problems of low efficiency of constant flow rate irrigation or damage to bile duct due to excessive pressure. S5. Use a long short-term memory network to make rolling predictions of future pressure and calculate the distance between image feature vectors before and after regulation to obtain the visual improvement rate; when the predicted pressure reaches the safety threshold or the visual improvement rate fails to meet the standard for a continuous period of time, trigger the second correction of the regulation parameters and the emergency pressure relief procedure. S5 includes the following steps: S51. Input the real-time pressure data into the long short-term memory network prediction model to calculate the pressure evolution trend line 500ms in the future. Specifically, S51 is as follows: the system is equipped with a lightweight LSTM (Long Short-Term Memory) network model, which takes the stress sequence of the past 2 seconds as input; the model outputs a stress prediction interval of 500ms for the next 500ms, with a confidence level of over 95%; Furthermore, the construction and training of the Long Short-Term Memory network prediction model includes the following steps: S511. Obtain the multidimensional sensing feature vector sequence, historical pump head speed execution sequence, and corresponding historical real intra-biliary pressure sequence from several historical cholangioscopic perfusion surgeries, and construct a historical pressure evolution dataset. S512. Based on the historical stress evolution dataset, construct an LSTM stress prediction model; S512 includes the following steps: S5121. Construct an initial LSTM stress prediction model and set a second training data ratio; divide the historical stress evolution dataset according to the second training data ratio to obtain a second training dataset and a second test dataset. S5122. Set a second training error threshold; input the second training dataset into the initial LSTM stress prediction model for training; during the training process, if the training error is less than the second training error threshold, stop training and obtain the trained LSTM stress prediction model; otherwise, continue training until the training error is less than the second training error threshold. S5123. Set a second test error threshold; input the second test dataset into the trained LSTM stress prediction model for testing; after the test is completed, obtain the second test error data; if the second test error data is less than or equal to the second test error threshold, use the trained LSTM stress prediction model as the final LSTM stress prediction model; otherwise, return to S5122 to continue training the trained LSTM stress prediction model and repeat S5123 until the second test error threshold is less than or equal to the second test error threshold. The second training data ratio, the second training error threshold, and the second test error threshold can be dynamically adjusted according to the model architecture, data volume, and actual needs during implementation. The training error and test error are both obtained by calculating the difference between the model output data and the benchmark label data using historical real stress sequences as the benchmark label through cosine similarity or Euclidean distance calculation. Furthermore, the initial LSTM pressure prediction model adopts a multi-LSTM model architecture comprising an input layer, two LSTM hidden layers, and a fully connected output layer. The time window length of the input layer is set to 2048 sampling points (corresponding to 2 seconds of data), and the feature dimensions include pressure value, gradient variance, high-frequency energy ratio, and PWM duty cycle of the inlet / outlet pumps. The first LSTM hidden layer contains 128 neurons, and the second LSTM hidden layer contains 64 neurons, used to extract long-short-term dependencies in the time series data. The fully connected output layer contains 512 neurons, corresponding to the pressure prediction value sequence within the next 500ms. S52. If the trend line touches the safety red line, immediately output a forced pressure reduction command, shut down the inlet pump and start the outlet pump at full speed. In specific implementation, S52 is as follows: once it is predicted that the pressure will reach 38cm H2O after 300ms (exceeding the safety red line of 30cm and leaving a margin), the system executes an emergency pressure relief command; at this time, the inlet pump solenoid valve is closed, and the outlet pump draws at maximum power (5000rpm) until the pressure drops back to 15cm H2O. S53. Set a step size deviation threshold, calculate the distance between the image feature vectors before and after adjustment as the visual improvement rate. If the visual improvement rate is lower than the step size deviation threshold within five consecutive adjustment cycles, issue a manual intervention prompt. S54. Store the successful parameter combination of this adjustment into the local memory and update the weight parameters of the strategy expert database online. The system employs a dual safety monitoring loop, combining LSTM pressure trend prediction and visual improvement rate assessment. Pressure prediction enables early warning of dangerous situations and emergency pressure relief response, transforming passive safety protection into proactive risk interception. Visual improvement rate assessment continuously monitors the effectiveness of regulation from the perspective of treatment effect, promptly prompting manual intervention when the flushing effect fails to meet the standard. At the same time, the expert database is continuously optimized through online learning of successful parameters, enabling the system to have a self-evolving ability that becomes more accurate with use.

[0024] For example, consider a patient with common bile duct stones and active bleeding undergoing surgery. At the start of the procedure, the surgeon inserts a rigid choledochoscope. Due to the stone fragmentation, the field of vision is obscured by red blood and gray stone powder, resulting in a visual impairment assessment value. V It quickly surged from 0.1 to 0.85; At this time, the system of the present invention captures this change in real time; the decoupling analysis in step S2 shows that the red staining masking degree is 0.7 and the particle suspension degree is 0.5; the step module immediately searches the expert database, determines that the current state is a high-risk fuzzy state, and generates a set of control parameters: requiring the execution of 4Hz biomimetic pulse flushing, setting the basic influent flow rate to 450ml / min, and the outfluent pump to perform equal volume follow-up. During step S4, the water pump generates sawtooth wave pulses, and the fluid impacts the bleeding points and debris piles like a miniature water hammer. Based on the feedback from step S432, the system detects that the field of view recovery is slow in the first two cycles, so it automatically increases the pulse duty cycle from 40% to 60%. At the 8th second of the operation, due to the sudden contraction of the patient's sphincter of O, the bile duct pressure prediction curve showed that it would exceed 40 cm H2O in 200 ms; step S52 intervened immediately, forcibly shutting off the water intake and suctioning, and the pressure was successfully cut off and dropped before touching the red line. Finally, at the 15th second, the visual clarity recovered to 90% of its original state, and the red staining basically disappeared; the system automatically switched to the stable laminar flow mode in step S44 to maintain the visual field; the entire process did not require the doctor to manually adjust the pressure switch, allowing the doctor to focus on the lithotripsy operation, improving surgical efficiency, and the intrabiliary pressure did not exceed the safety threshold throughout the process, effectively avoiding biliary infection; this example demonstrates the ability of the present invention to achieve intelligent and precise control through dual feedback of vision and pressure in the complex and ever-changing intrabiliary environment.

[0025] For further details, please refer to Figure 5 A rigid cholangioscopy perfusion pressure control system based on field-of-view optimization is provided to implement the aforementioned rigid cholangioscopy perfusion pressure control method based on field-of-view optimization. The system includes: The multidimensional data acquisition module is used to acquire and preprocess intraductal pressure signals and video image streams in real time, and output pressure time series, image sequence to be analyzed and multidimensional sensing feature vectors. The visual feature analysis module receives the image sequence to be analyzed and the stress time series, decouples the causes of visual blurring and calculates the comprehensive evaluation value of visual impairment; The control decision center module generates the optimal control parameter set based on the comprehensive evaluation value of visual obstruction, the classification of ambiguity causes, and the current pressure. The adaptive execution drive module drives the inlet and outlet water pumps to generate a coordinated biomimetic pulse flow field based on the optimal control parameter set and feeds back the data after execution. The safety monitoring and feedback module uses stress prediction and the visual improvement rate calculated from the multi-dimensional perception feature vector to perform closed-loop safety monitoring and strategy updates. This invention abandons the passive mechanical mode of traditional endoscopic perfusion systems, which rely on single pressure feedback and constant flow rate irrigation. Instead, it proposes a biomimetic closed-loop architecture that integrates visual perception, fluid decision-making, and predictive execution. It combines the decoupling of depth features from computer vision with pulse control from fluid dynamics and introduces a time-series prediction model. The system automatically applies customized turbulence or pulses to different impurities such as blood, gas, and stones, improving debridement efficiency and eliminating the tedious manual parameter adjustments required by physicians. A dual-pump synergy and forced pressure relief mechanism based on prediction strictly locks the intrabiliary pressure within physiological safety thresholds, eliminating the risk of iatrogenic bile duct injury. The system possesses the ability to correct and optimize based on visual improvement rates, providing an absolutely safe, clear, and stable intelligent operating environment for complex minimally invasive hepatobiliary surgeries.

[0026] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0027] The preferred embodiments of the invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention.

Claims

1. A method for controlling perfusion pressure in rigid cholangioscopy based on blurred field of view optimization, characterized in that, Includes the following steps: S1. Simultaneously acquire the original pressure signal and original video stream inside the bile duct in real time, perform multi-scale feature preprocessing on the original video stream to generate an image sequence to be analyzed containing spatial gradient and frequency domain distribution features, and extract multi-dimensional sensing feature vectors at the same time. S2. Input the image sequence to be analyzed and the multidimensional perceptual feature vector into the visual blurring cause decoupling model, combine it with the current pressure signal, quantify the red staining degree, particle suspension degree, bubble interference degree and tissue adhesion degree in real time, and calculate the comprehensive evaluation value of visual impairment accordingly. S3. Based on the comprehensive evaluation value of visual obstruction, the quantitative results of each ambiguity cause, and the current pressure value, retrieve the perfusion strategy expert database and generate a preliminary perfusion control parameter set including flushing intensity, fluid replacement rate, and target pressure range. S4. Based on the preliminary infusion control parameter set, the coordinated speed curve of the inlet pump and the outlet pump is generated by adaptive pulse width modulation, and the biomimetic pulse fluid is output to perform impurity stripping and discharge, and the feedback pressure and real-time image after execution are obtained. S5. Use a long short-term memory network to make rolling predictions of future pressure and calculate the distance between image feature vectors before and after regulation to obtain the visual improvement rate; when the predicted pressure reaches the safety threshold or the visual improvement rate fails to meet the standard for a continuous period of time, trigger the secondary correction of regulation parameters and emergency pressure relief procedure.

2. The method for controlling perfusion pressure in rigid cholangioscopy based on visual field ambiguity optimization according to claim 1, characterized in that, S1 includes the following steps: S11. Using a piezoresistive miniature pressure sensor integrated into the tip of the rigid cholangioscope sheath, the instantaneous pressure in the bile duct is collected, and the baseline drift is removed by median filtering to obtain a smooth pressure time series. S12. Obtain the original video stream through the photoelectric conversion module, calculate the global gradient variance of each frame image using the Laplacian operator, and use it as the initial sharpness reference value to complete the initial image quality judgment. S13. After the initial quality judgment, the image is grayscaled and histogram equalized to enhance the edges of debris and bubbles. The high-frequency component energy ratio is extracted by fast Fourier transform, and the image is subjected to local contrast analysis to mark suspected blur candidate areas. S14. Align the global gradient variance, high-frequency energy ratio, suspected fuzzy candidate region labeling information with the smoothed pressure signal along the time axis to construct a multi-dimensional perceptual feature vector and simultaneously form an image sequence to be analyzed.

3. The method for controlling perfusion pressure in a rigid cholangioscope based on visual field ambiguity optimization according to claim 2, characterized in that, The step of marking suspected fuzzy candidate regions in S13 includes the following steps: S131. Set a pixel sliding window, traverse the images in the original video stream based on the pixel sliding window, calculate the local contrast factor, and mark the areas with contrast below the threshold as suspected blur candidate areas. S132. Set a global average density threshold; perform Sobel edge detection on suspected blurry candidate areas, and count the edge pixel density. If the edge density of a certain area is lower than the global average density threshold, it is determined that there is substantial visual occlusion in the area, and the determination information is incorporated into the multidimensional perception feature vector.

4. The method for controlling perfusion pressure in a rigid cholangioscope based on visual field ambiguity optimization according to claim 2, characterized in that, S2 includes the following steps: S21. Input the image sequence to be analyzed into the visual blurring cause decoupling model, and obtain the red occlusion degree by red channel pixel ratio analysis. S22. Use morphological closing operations to identify circular or elliptical bright patches in the image, calculate the number of patches and the movement vector, and obtain the bubble interference degree. S23. Analyze the high-frequency noise distribution pattern of the image, identify suspended particles with a diameter smaller than a preset threshold and calculate their coverage area ratio to obtain the particle suspension degree. S24. Combining the fluctuation characteristics of the pressure time series, when the pressure suddenly increases and a large area of ​​low texture appears on the screen, determine the tissue adhesion and calculate the tissue adhesion degree. S25. The red staining opacity, bubble interference, particle suspension and tissue adhesion are weighted and fused to obtain a comprehensive evaluation value of visual obstruction.

5. The method for controlling perfusion pressure in a rigid cholangioscope based on visual field ambiguity optimization according to claim 4, characterized in that, The process of obtaining tissue adhesion in step S24 includes the following steps: S241. Extract the mutual information entropy of ten consecutive frames of images. When the slope of the mutual information entropy decreases exceeds the threshold and the pressure signal shows increased resistance, start tissue adhesion-assisted recognition. S242. Calculate the motion vector in the central area of ​​the image using the optical flow algorithm. If the vector length approaches zero and the pixel grayscale distribution is highly consistent, then it is confirmed that the lens is completely blocked by the wall of the bile duct or a large piece of tissue.

6. The method for controlling perfusion pressure in rigid cholangioscopy based on visual field ambiguity optimization according to claim 4, characterized in that, S3 includes the following steps: S31. Calculate the minimum flushing volume required to replace the original fluid in the bile duct based on the red staining degree, and use it as the basic flow compensation value of the inlet pump. S32. Based on the particle suspension degree and bubble interference degree, match the corresponding flow field disturbance mode from the strategy expert library, including laminar displacement mode, turbulent scouring mode and intermittent negative pressure attraction mode. S33. Set the target pressure range and calculate the pressure adjustment ratio coefficient based on the deviation between the current pressure and the target range; S34. Couple the basic flow compensation value, the matched flow field disturbance mode and the pressure regulation proportional coefficient to generate a preliminary set of control parameters including pump head speed, start-stop cycle and the difference in flow between internal and external pumps. S35. Introduce an environmental compliance factor that reflects the degree of bile duct dilation to dynamically correct the flow gain in the initial control parameter set.

7. The method for controlling perfusion pressure in a rigid cholangioscope based on visual field ambiguity optimization according to claim 6, characterized in that, S4 includes the following steps: S41. The preliminary control parameter set is converted into the PWM duty cycle signal of the water inlet pump controller to control the water inlet pump to generate sawtooth wave pressure pulses and use fluid impact force to peel off the adhering impurities. S42. Set the water flow rate multiple; synchronously adjust the suction power of the outlet pump to keep the outlet flow rate at a multiple of the inlet flow rate, so as to maintain the bile duct lumen filling stably; S43. Monitor the pressure change rate in the bile duct in real time after execution, and adjust the duty cycle of the pulse waveform through the feedback loop; S44. Set the red dyeing masking threshold; when the red dyeing masking is detected to drop below the red dyeing masking threshold, switch the pulse flushing to the steady laminar flow mode.

8. The method for controlling perfusion pressure in a rigid cholangioscope based on visual field ambiguity optimization according to claim 7, characterized in that, The feedback loop adjustment in S43 includes the following steps: S431. Calculate the ratio of the current pressure change rate to the safe pressure change threshold. If the ratio is greater than 1, reduce the amplitude of the water inlet pulse proportionally. S432. Calculate the slope of the improvement in visual clarity over the past three pulse cycles. If the slope is less than the optimization target value, gradually increase the pulse frequency within the safety pressure limit. The slope of the improvement in visual clarity is calculated from the gradient variance evolution extracted in S1.

9. The method for controlling perfusion pressure in a rigid cholangioscope based on visual field ambiguity optimization according to claim 7, characterized in that, S5 includes the following steps: S51. Input real-time pressure data into the long short-term memory network prediction model to calculate the pressure evolution trend line at future moments; S52. If the trend line touches the safety red line, immediately output a forced pressure reduction command, shut down the inlet pump and start the outlet pump at full speed. S53. Set a step size deviation threshold, calculate the distance between the image feature vectors before and after adjustment as the visual improvement rate. If the visual improvement rate is lower than the step size deviation threshold within five consecutive adjustment cycles, issue a manual intervention prompt. S54. Store the successful parameter combination of this adjustment into the local memory and update the weight parameters of the strategy expert database online.

10. A rigid cholangioscope perfusion pressure control system based on field-of-view ambiguity optimization, characterized in that, For implementing the rigid cholangioscopy perfusion pressure control method based on visual field ambiguity optimization as described in any one of claims 1-9, the system comprises: The multidimensional data acquisition module is used to acquire and preprocess intraductal pressure signals and video image streams in real time, and output pressure time series, image sequence to be analyzed and multidimensional sensing feature vectors. The visual feature analysis module receives the image sequence to be analyzed and the stress time series, decouples the causes of visual blurring and calculates the comprehensive evaluation value of visual impairment; The control decision center module generates the optimal control parameter set based on the comprehensive evaluation value of visual obstruction, the classification of ambiguity causes, and the current pressure. The adaptive execution drive module drives the inlet and outlet water pumps to generate a coordinated biomimetic pulse flow field based on the optimal control parameter set and feeds back the data after execution. The safety monitoring and feedback module uses the visual improvement rate calculated from the stress prediction and the multi-dimensional perception feature vector to perform closed-loop safety monitoring and strategy updates.

Images