An automatic suction sputum bronchoscope system for airway secretion clearance

Abstract

This invention discloses an automated bronchoscope system for clearing airway secretions, relating to the field of automated bronchoscopy. The system includes: an endoscope image acquisition module for real-time acquisition of images of the airway lumen distal to the bronchoscope; a secretion recognition module for analyzing the airway lumen images to identify secretion targets and output their aspirability characteristics; an operation decision module for determining, based on the aspirability characteristics, whether the current secretion target is within a preset airway level where suction can be performed, and generating a suction command if the determination is positive; and a suction execution module for applying negative pressure to the secretion target through the suction channel of the bronchoscope in response to the suction command. This solution achieves precise positioning and selective suction of sputum targets within the airway, effectively avoiding the risk of aspirating airway tissue or foreign objects, and improving the success rate of clearing secretions of different characteristics.

Description

Technical Field

[0001] This invention relates to the field of automated bronchoscopy, and particularly to an automated suction bronchoscopy system for clearing airway secretions. Background Technology

[0002] In the fields of intensive care, neurological rehabilitation, and geriatric medicine, the number of patients with airway secretion retention due to weakened or absent cough reflexes is increasing. These patients are unable to effectively clear sputum from their airways independently, and long-term accumulation of secretions can easily lead to atelectasis, recurrent infections, and even respiratory failure. Clinically, multiple bedside fiberoptic bronchoscopic suctioning procedures per day are usually required to maintain airway patency. However, existing bronchoscopic technology systems are mainly built around diagnostic and treatment scenarios such as tumor biopsy, complex interventional therapy, and 3D navigation based on CT images. Their design goal is to assist doctors in performing high-precision and high-difficulty operations, rather than solving basic airway clearance needs. Although image recognition and robot-assisted technologies have been gradually introduced into the field of respiratory endoscopy in recent years, existing solutions are still mainly semi-automatic, lacking a systematic solution that can autonomously complete secretion identification and suctioning decisions based on real-time endoscopic vision.

[0003] Current bronchoscopic suctioning procedures rely entirely on manual operation, which has the following technical limitations: First, the effectiveness of the procedure is highly dependent on the physician's personal experience and operational stability, making it difficult to guarantee standardization for each procedure, especially for patients requiring repeated daily suctioning. Second, existing auxiliary technologies primarily focus on preoperative CT-based three-dimensional airway reconstruction and path planning, addressing navigation issues but failing to autonomously identify secretions appearing in real-time during the procedure. Third, existing systems lack safety monitoring mechanisms for the suctioning procedure itself, failing to promptly stop the procedure and provide intervention in cases of sputum blockage, ineffective suctioning, or abnormal propulsion resistance, posing potential operational risks. Fourth, the lack of a systematic record-keeping system for historical suctioning paths and cleared bronchial segments in clinical practice leads to the potential omission of airway branches or ineffective repetition of cleared areas during repeated procedures, reducing the overall efficiency of airway management. Therefore, based on these challenges, this invention proposes an automated bronchoscope suctioning system for airway secretion removal. Summary of the Invention

[0004] To address the aforementioned problems, the present invention aims to provide an automated sputum suction bronchoscopic system for clearing airway secretions. By introducing a secretion recognition and aspirability assessment mechanism based on real-time endoscopic images, the system enables autonomous identification and negative pressure suction of sputum targets within a preset airway range. Simultaneously, the built-in safety monitoring and path recording module ensures the safety and traceability of the automated sputum suction process.

[0005] To achieve the above objectives, this invention provides an automated suction bronchoscopy system for clearing airway secretions. The system comprises: a secretion identification module that analyzes real-time intraluminal airway images and outputs suctionability parameters of the target secretion; an operation decision module that determines whether to perform suction based on these parameters and the current airway level, limiting the operation to a preset area from the main bronchus to the segmental bronchus; a suction execution module that responds to the decision command by applying negative pressure suction through the suction channel and adaptively adjusting suction parameters based on the dynamic image characteristics of the secretions; a safety monitoring module that interrupts the automated operation and prompts for manual intervention when it detects ineffective suction, insufficient image recognition confidence, or abnormal negative pressure response; and an operation path recording module that constructs an airway topology based on a continuous image sequence, marks cleared branches, and guides the sequential completion of the cleaning task.

[0006] In a first aspect, the present invention provides an automated suction bronchoscope system for clearing airway secretions, comprising: The endoscopic image acquisition module is used to acquire real-time images of the airway lumen distal to the bronchoscope. The secretion recognition module is used to analyze the images inside the airway cavity to identify secretion targets and output their aspirability characteristic parameters; The operation decision module is used to determine whether the current secretion target is within the preset airway level range where suction operation can be performed based on the suctionability characteristic parameters, and to generate a suction command when the determination is affirmative. A suction execution module is used to apply negative pressure to the target secretion through the suction channel of the bronchoscope in response to the suction command; The safety monitoring module is used to monitor the status parameters of the attraction process in real time, and to interrupt the automatic operation and issue a manual takeover request when an abnormal state is detected. The operation path recording module is used to identify bronchial bifurcation markers based on a continuous sequence of endoscopic images, construct the topology of the current operation path, and mark the bronchial branches for which suction operations have been performed.

[0007] Furthermore, the system is suitable for patients who require long-term mechanical airway clearance due to weakened or absent cough reflexes.

[0008] Furthermore, the secretion recognition module distinguishes mobile sputum secretions from fixed airway tissues or foreign bodies based on at least one of the morphological features, light reflection characteristics, or fluidity characteristics of secretions in optical images.

[0009] Furthermore, the preset airway hierarchy range for suction operations includes the main bronchus, lobar bronchus, and segmental bronchus. The operation decision module identifies the current hierarchy based on the anatomical morphological characteristics of bronchial bifurcation and prohibits automatic suction of terminal bronchus outside this range, so as to ensure that the operation is limited to the preset safe area and avoid excessive intrusion or damage to the distal airway.

[0010] Furthermore, the operation decision module is also used to generate advancement navigation prompts for uncleared branches based on the cleared branch information provided by the operation path recording module, so as to guide the bronchoscope to complete the clearing task within the preset airway range in sequence, thereby avoiding repeated operations on cleared areas or omissions of uncleared areas, and improving the efficiency and completeness of clearing.

[0011] Furthermore, the suction execution module adaptively adjusts the intensity parameters of negative pressure suction based on the local geometry of the airway where the secretion is located, including the airway diameter and the angle between the suction direction and the airway centerline, combined with the spatial volume characteristics of the secretion target, so that the effective suction force applied to the secretion target matches its attachment state.

[0012] Furthermore, the safety monitoring module includes a multimodal sensor fusion unit for simultaneously acquiring endoscopic image features, contact force features between the bronchoscope tip and the airway wall, and fluid dynamics features within the suction channel; the safety monitoring module identifies abnormal states during the suctioning process based on the collaborative analysis of multimodal features, and executes a graded interruption response according to the overall severity of the abnormality.

[0013] Furthermore, the abnormal states include the secretion target not being cleared after a preset number of suction operations, the confidence level of the secretion recognition module in recognizing the target in the current field of view being lower than a threshold, or the negative pressure response curve detected by the suction execution module deviating from the preset fluid dynamic characteristics. By monitoring these states, a multi-dimensional safety assessment of the suction process can be achieved, ensuring that the operation is stopped in time when an abnormality occurs.

[0014] Furthermore, after issuing a manual takeover request, the security monitoring module switches system control to manual operation mode and continues to monitor the status of subsequent operations until manual confirmation is received to restore automatic mode.

[0015] In a second aspect, an automated suction bronchoscopy method for clearing airway secretions is also provided, the method being based on the system described in the first aspect above, comprising: Real-time acquisition of images of the airway lumen distal to the bronchoscope; The images inside the airway are analyzed to identify the target secretion and output its aspirability characteristics. Based on the aspirability characteristic parameters, determine whether the current secretion target is within the preset airway level range where aspiration can be performed; When the determination is positive, a suctioning instruction is generated and negative pressure is applied to the target secretion through the suction channel of the bronchoscope; The system monitors status parameters in real time during the attraction process. When an abnormal state is detected, the automatic operation is interrupted and a manual takeover request is issued.

[0016] This invention provides an automated bronchoscope system and method for clearing airway secretions. Its core features include: a real-time airway image acquisition module that uses an endoscope to acquire images of the airway lumen; a secretion recognition module that analyzes image features to output suctionability parameters of the target secretion; an operation decision module that determines whether to perform suction within a preset range from the main bronchus to the segmental bronchus based on these parameters and the current airway level identified by bronchial bifurcation morphology, generating a suction command if the condition is met; a suction execution module that responds to the command by applying negative pressure through the suction channel, while adaptively adjusting suction parameters based on dynamic image features of the secretions; a safety monitoring module that monitors the status parameters of the suction process in real time, interrupting automatic operation and prompting manual intervention when suction is ineffective, recognition confidence is insufficient, or negative pressure response is abnormal; and an operation path recording module that constructs an airway topology based on a continuous image sequence, marks cleared branches, and guides the sequential completion of the cleaning task.

[0017] This solution introduces a mechanism for autonomous secretion identification and aspirability assessment based on real-time endoscopic images, enabling precise localization and selective aspiration of sputum targets within the airway. This effectively avoids the risk of aspiration of airway tissue or foreign bodies and significantly improves the success rate of clearing secretions of different characteristics. By limiting the operation range to preset airway levels and incorporating safety monitoring and abnormal interruption functions based on multi-dimensional status parameters, the risk of excessive intrusion into the distal airway during automated operation is fundamentally reduced, ensuring the safety of the operation process. Through the automatic construction of the operation path and the marking of cleared areas, the traceability and systematic nature of the cleaning process are achieved, avoiding repetitive operations or omissions and improving the overall efficiency and integrity of airway management.

[0018] Beneficial effects By implementing the automated sputum suction bronchoscope system for clearing airway secretions provided by the present invention, the following technical effects are achieved: (1) By automatically identifying bronchial bifurcation markers in a continuous endoscopic image sequence and determining the current airway level in real time, the automated suctioning operation is strictly limited to a safe area within the preset main bronchus to segmental bronchus. This effectively avoids excessive intrusion of the distal small bronchi by automated suctioning, reduces the risk of tissue damage, airway spasm, and perforation caused by deep manipulation, and ensures that the automated suctioning process always operates within the anatomically safe boundary, providing a basic safety guarantee for repeated cleaning operations.

[0019] (2) By constructing the topology of airway branches and marking the bronchial segments that have undergone suction operations in real time, the system generates propulsion navigation prompts for uncleaned areas, guiding the bronchoscope to complete a comprehensive cleaning within a preset range in a predetermined order. This mechanism achieves traceability and systematicity in the cleaning process, ensuring that each operation can completely cover the target airway area, avoiding ineffective duplication of work and the generation of cleaning blind spots, and significantly improving the overall efficiency of airway management and the integrity of the cleaning process.

[0020] (3) By introducing quantitative evaluation indicators based on the dynamic deformation characteristics of secretions, interfacial optical properties, and optical contrast, the system achieves graded identification and suction level classification of secretions with different properties. This significantly improves the selective removal success rate of viscous and highly adhesive secretions, while effectively reducing the number of ineffective suctions caused by misjudgment. This makes the automated suctioning operation more targeted and adaptable, avoiding repeated attempts due to insufficient suction or mucosal irritation caused by excessive suction.

[0021] (4) By reconstructing the local geometry of the airway containing secretions in real time and combining it with the volume characteristics of the secretions, the optimal negative pressure value matching the current anatomical conditions is dynamically calculated. This achieves precise matching between negative pressure intensity and local airway characteristics: while maintaining efficient clearance in the main airway area, the negative pressure amplitude is automatically reduced in the distal narrow airway, effectively suppressing the negative pressure amplification effect caused by local airway narrowing, and significantly reducing excessive negative pressure impact on the distal bronchial mucosa, thereby minimizing the risk of airway damage while ensuring clearance effect.

[0022] (5) By simultaneously fusing endoscopic image features, contact force characteristics between the bronchoscope tip and the airway wall, and fluid dynamics characteristics within the suction channel, a collaborative monitoring system covering optical, mechanical, and fluid dimensions was constructed. This system accurately identifies and classifies various complex abnormal states, such as abnormal airway wall contact, progressive blockage of the suction channel, and image recognition failure, and executes graded interruption responses based on the overall severity of the abnormality. This method significantly improves the detection rate of abnormal events during automated suctioning, shortens the response delay from the occurrence of an abnormality to manual intervention prompts, and ensures the safety of the operation process from multiple dimensions. Attached Figure Description

[0023] To make the above-described automated suction bronchoscopy system for clearing airway secretions of the present invention more apparent and understandable, the accompanying drawings used in the specific embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below 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.

[0024] Figure 1 This is a flowchart illustrating the method described in this application; Figure 2 This is a flowchart illustrating the automated suctioning process. Detailed Implementation

[0025] Example 1: This embodiment provides an automated suction bronchoscopy system and method for clearing airway secretions. The method flow is as follows: Figure 1 As shown, the system includes an endoscopic image acquisition module, a secretion recognition module, an operation decision module, a suction execution module, a safety monitoring module, and an operation path recording module. These modules work together to form a complete technical closed loop.

[0026] Automated suctioning process as follows Figure 2 As shown, after the system is started and enters the airway, the endoscopic image acquisition module acquires images of the airway lumen distal to the bronchoscope in real time at a preset frequency, providing basic data for subsequent analysis. The secretion recognition module first performs preliminary detection and segmentation of the secretion targets in the images, extracting their morphological features, light reflection characteristics, and edge gradient information. To further evaluate the actual absorbability of the secretions, the system introduces a dynamic evaluation mechanism based on the rheological properties of the secretions and their adhesion to the airway wall. Specifically, the control module instructs the suction execution module to apply a weak airflow pulse through the suction channel. The pulse intensity is insufficient to aspirate the secretions, but sufficient to deform the relatively fluid sputum. The secretion recognition module continuously acquires image sequences before and after the disturbance, and by analyzing the relative change rate of the secretion's projected area, the change in edge grayscale gradient, and the optical contrast between the secretions and the airway wall, it comprehensively generates a quantitative absorbability evaluation index, and classifies the secretions into different absorbability levels accordingly. Thin sputum typically exhibits high deformability and low optical contrast, and is classified as highly absorbable; viscous sputum plugs have moderate deformability and a high edge gradient, and are classified as moderately absorbable; while dry crusts highly adhered to the airway wall show almost no deformation, have sharp edges and low contrast with the tissue, and are classified as low absorbable. This classification mechanism provides a basis for subsequent implementation of differentiated suction strategies, avoiding ineffective or excessive suction caused by using uniform operating parameters for secretions of different characteristics.

[0027] Furthermore, to ensure robust operation of the system in complex airway environments, this system also incorporates an image quality assessment mechanism. This mechanism operates in parallel with the secretion recognition module, generating an image quality index by analyzing the clarity, contrast, illumination uniformity, and presence of large-area obstructions in real time. When the image quality index falls below a first preset threshold, the system prioritizes triggering a self-cleaning process, such as applying short-duration high-pressure gas or saline pulses through the suction channel or auxiliary airway to attempt to remove deposits from the lens surface. If the image recovers to a satisfactory level after self-cleaning, the system continues to perform automatic secretion recognition and suction operations; if the image quality still does not improve, the system will automatically lower the confidence requirement for secretion recognition, performing suction only on sputum targets with extremely obvious features, suspending automatic judgment of fine anatomical layers, switching the operation mode to semi-automatic, and awaiting operator confirmation. When the image quality index falls below the second preset threshold (i.e., the image is severely damaged, such as being completely obscured by blood), the system immediately considers it a serious abnormal state, interrupts all automatic operations, issues a vision loss alarm through the safety monitoring module, automatically retracts the bronchoscope to the previous safe area, and forcibly switches to manual control mode until the operator confirms that the vision has been restored and manually unlocks the system.

[0028] Before generating a suctioning command, the operation decision module first automatically identifies bronchial bifurcation landmarks based on a continuous endoscopic image sequence. By analyzing the bifurcation morphology and airway direction, it determines the anatomical level of the bronchoscope tip in real time and strictly limits the automatic suctioning operation to the main bronchus, lobar bronchus, and segmental bronchus. If the current level exceeds the preset executable range, the system immediately stops the automatic suctioning process and outputs a prompt message, guiding the operator to return to a safe area or wait for manual confirmation. If the system has an automatic withdrawal function, it can automatically withdraw the scope to the previous safe bifurcation point while recording the current position for subsequent processing.

[0029] After confirming that the current level is operational, the system further assesses the attractability level of the secretion target. For targets with attractability below a set threshold, the system will not perform automatic aspiration, but will mark the target in the image and record its location and characteristics for subsequent manual intervention. At the same time, the system can choose to attempt a second assessment or skip the target and continue to the next area.

[0030] For targets that are highly attractive but located in an inoperable zone, the system strictly adheres to the safety-first principle, prohibiting automatic suctioning and alerting the operator that the target is in a high-risk area, suggesting manual assessment of the appropriate treatment. This mechanism ensures that automated suctioning operations always operate within preset safety boundaries, balancing both safety and effectiveness.

[0031] Once the secretions are determined to be within an operable range, the suction execution module prepares to perform suctioning. To adapt to the varying suction requirements of different airway locations, an adaptive negative pressure adjustment mechanism based on local airway geometry is proposed. Before suctioning, the system controls the bronchoscope tip to perform a micro-scanning motion, reconstructing the local three-dimensional geometry of the airway in the secretion area using multi-view images, extracting the local airway diameter and the angle between the suction channel outlet axis and the airway centerline. Simultaneously, the spatial volume of the secretions is estimated based on image segmentation and three-dimensional reconstruction. Based on the aforementioned geometric parameters and volume characteristics, the suction execution module dynamically calculates the optimal negative pressure value matching the current anatomical conditions: for larger areas such as the main airway, the negative pressure value is close to the conventional setting level to ensure efficient removal; for smaller areas such as segmental bronchi, the system automatically reduces the negative pressure amplitude while increasing the suction duration or using a pulsed suction mode to compensate for the local negative pressure amplification effect caused by airway narrowing. This mechanism enables precise matching between the effective suction applied to the secretion target and its attachment state and location, minimizing excessive negative pressure impact on the airway wall while ensuring effective removal.

[0032] Throughout the automated suctioning procedure, the safety monitoring module operates continuously, monitoring the operational status from multiple dimensions. The system employs multimodal sensor fusion technology to simultaneously acquire three types of characteristic information: first, optical features extracted from endoscopic images, such as the degree of mucosal deformation, mucosal congestion index, and confidence level of secretion identification; second, the contact force vector between the endoscope and the airway wall detected by a miniature fiber optic force sensor array integrated into the bronchoscope's tip; and third, the negative pressure value and airflow rate monitored in real-time by pressure and flow sensors at the suction channel inlet, along with fluid dynamic characteristics calculated from these, such as pressure build-up time and steady-state pressure fluctuation range. Based on the collaborative analysis of these three types of features, the safety monitoring module identifies abnormal states during the suctioning process. When any dimension exceeds a preset safety threshold, the system determines it to be abnormal: significant mucosal deformation or a sudden drop in identification confidence indicates an optical dimension abnormality; excessive contact force amplitude or continuous pointing towards the airway wall indicates a mechanical dimension abnormality; and excessively long pressure build-up time or a pressure-flow curve exhibiting a blockage pattern indicates a fluid dimension abnormality. Based on the number and severity of anomalies, the system implements a tiered interruption response: a warning is issued for minor anomalies in a single dimension, but operation is not interrupted; when two dimensions are simultaneously abnormal or a single dimension is persistently abnormal, operation is immediately suspended while maintaining the scope's position, and a manual intervention prompt is output; in the event of an emergency anomaly, the scope withdrawal procedure is automatically executed and an emergency intervention request is issued. This multi-dimensional collaborative monitoring mechanism significantly improves the accuracy of identifying various anomalies and the timeliness of response.

[0033] After one or more suctioning procedures are completed, the procedure path recording module identifies the morphological features of each bronchial bifurcation based on a sequence of continuous endoscopic images, constructs the topology of the current procedure path, and marks the bronchial branches that have undergone suctioning in real time. Based on the relative positions of the cleared and uncleaned branches, the system generates navigation prompts for the uncleaned areas, guiding the bronchoscope to complete a comprehensive cleansing within a preset range. This mechanism avoids repeated cleaning or omissions due to operator fatigue or memory lapses, ensuring that each procedure systematically covers the target airway area, achieving traceability and completeness of the cleaning process.

[0034] Example 2: Building upon the aforementioned embodiments, and addressing the limitation of existing methods that can only identify the presence of secretions but cannot quantify their attractability, a dynamic evaluation method based on the rheological properties of secretions and their adhesion to the airway wall is proposed. This method constructs a joint evaluation model of secretion viscoelasticity and adhesion by analyzing the deformation characteristics, edge fluidity, and optical interface properties between secretions and the airway wall during airflow disturbances or slight endoscope displacement. When secretions are subjected to weak airflow pulses or slight movements of the endoscope tip, the degree of morphological change is functionally related to their viscosity, surface tension, and adhesion strength. By introducing a secretion attractability index characterizing the dynamic response of secretions, the deformation feature parameters extracted from the image sequence are mapped to a quantified attractability score, providing the control module with a more refined operational decision-making basis. This enables a graded attraction strategy for secretions with different properties, avoiding ineffective operations due to insufficient attraction or airway mucosal damage due to excessive attraction.

[0035] After the bronchoscope reaches the target airway, the control module first instructs the endoscope image acquisition module to continuously acquire no less than 3 frames of static images at a first sampling frequency. After image registration and denoising, a reference image of the secretion target is generated. Simultaneously, the image recognition module segments the secretion region, calculates its initial projected area, and extracts the mean gray-level gradient of the secretion edge region, as well as the average light intensity of the edge region and the average light intensity of the adjacent airway wall region.

[0036] The control module instructs the suction execution module to apply a weak negative pressure pulse of preset duration and amplitude through the suction channel of the bronchoscope, or a weak positive pressure pulse through the auxiliary airway built into the bronchoscope. The intensity of this pulse is insufficient to draw secretions into the suction channel, but sufficient to cause observable deformation or displacement of loosely attached secretions. During pulse application, the endoscopic image acquisition module continuously acquires dynamic image sequences of the secretion area at a second sampling frequency.

[0037] The image recognition module analyzes the acquired dynamic image sequence, tracks the displacement trajectory of the secretion edge using inter-frame difference and optical flow methods, calculates the maximum projected area of ​​the secretion after disturbance or the projected area after reaching steady state, and then obtains the relative rate of change of the projected area. Simultaneously, it recalculates the mean grayscale gradient and light intensity ratio of the secretion edge region and compares them with the baseline values.

[0038] The formula for calculating the secretion absorbability index (SAI) is as follows:

[0039] In the formula, These are preset weighting coefficients used to adjust the contribution of liquidity characteristics to attractiveness assessment; It is the relative rate of change of the projected area of ​​secretions after being subjected to an airflow pulse of a preset amplitude, reflecting the fluidity of secretions under the action of external force; These are preset weighting coefficients used to adjust the contribution of interface gradient features to attractiveness assessment. The mean gray-scale gradient of the secretion edge region is used to characterize the clarity of the interface between the secretion and the airway wall. The more tightly the secretion is attached, the higher its edge gradient value. These are preset weighting coefficients used to adjust the contribution of optical contrast features to attractiveness assessment; It is the ratio of the average light intensity of the edge region of the secretion to the average light intensity of the adjacent airway wall region, used to quantify the optical contrast between the secretion and the airway wall.

[0040] Among them, the weighting coefficient , , The settings can be pre-trained using machine learning methods based on clinical sample data, or manually set by the operator through a human-computer interaction terminal according to the patient's specific condition.

[0041] The operation decision module generates a tiered attraction strategy based on the calculated SAI value and a preset threshold range. If the SAI is higher than the first threshold, such as 0.7, it is determined to be highly absorbable secretion, and standard negative pressure suction is performed directly. If the SAI is between the first and second thresholds, such as 0.3 to 0.7, it is determined to be moderately absorbable secretion, and enhanced suction operation is performed, including extending the suction duration, increasing the negative pressure amplitude, or applying pulsed suction. If the SAI is below the second threshold, such as 0.3, it is determined to be low-absorbability secretion. The system will not perform automatic aspiration, but will record the location and output a prompt, suggesting manual intervention or other physical cleaning methods.

[0042] After each suction operation, the system recalculates the SAI value of the secretion area, evaluates the suction effect, and adaptively adjusts the parameters of the next suction operation based on the effect, forming a closed-loop feedback control.

[0043] By introducing a dynamic aspirability assessment method based on the rheological properties of secretions and their adhesion to the airway wall, precise graded aspiration of secretions with different characteristics was achieved. Compared with conventional image recognition technology, improvements were made in secretion clearance success rate and airway mucosal protection. The comparative experiment was set up as follows: 60 patients receiving invasive mechanical ventilation were randomly divided into a control group (using conventional image recognition + fixed negative pressure aspiration) and an experimental group (using SAI index assessment + graded aspiration strategy). In the experimental group, secretions with an SAI value higher than 0.7 were aspirated using standard negative pressure, secretions with an SAI value between 0.3 and 0.7 were aspirated using enhanced aspiration, and secretions with an SAI value lower than 0.3 had automatic aspiration paused and were prompted for manual intervention. The experimental results showed that the success rate of clearing thin sputum in a single attempt was 94.3% in the experimental group, an increase of 20.6% compared to 78.2% in the control group; the success rate of clearing viscous sputum plugs was 81.7%, an increase of 55.9% compared to 52.4% in the control group; and the aspiration rate of highly adhesive dry crusts decreased from an average of 2.3 times per case in the control group to 0.4 times per case in the experimental group, a reduction of 82.6%. These results indicate that by quantitatively assessing the rheological properties of secretions and implementing a graded aspiration strategy, the targeting and effectiveness of clearing secretions of different characteristics can be significantly improved, while effectively avoiding ineffective operations on non-aspirable targets.

[0044] Example 3: Building upon the aforementioned embodiments, this paper addresses the shortcomings of existing methods that employ fixed negative pressure parameters during suctioning and cannot differentiate them based on the specific geometry of the airway where the secretions are located. An adaptive negative pressure control method based on airway geometry and fluid dynamics simulation is proposed. In this method, the shear force experienced by the secretions during suction depends not only on the applied negative pressure value but also on the relative position between the suction channel inlet and the secretions, the airway diameter, the branching angle, and the rheological properties of the secretions themselves. The actual suction force experienced by the secretions differs significantly when the same negative pressure is applied to the main bronchus versus the segmental bronchus. By reconstructing the local three-dimensional geometry of the airway where the secretions are located in real time and combining it with a fluid dynamics model, the theoretical minimum negative pressure value required to effectively detach the secretions from their attachment location and draw them into the suction channel is calculated, thereby achieving precise adaptive adjustment of the negative pressure parameters.

[0045] Once the image recognition module detects the secretion target, the control module instructs the bronchoscope tip to perform a preset scanning motion, while the endoscope image acquisition module acquires multi-view image sequences at a high frame rate. Based on motion reconstruction structures or monocular vision depth estimation methods, the image recognition module reconstructs a three-dimensional point cloud model of the local airway at the location of the secretion, and then extracts geometric parameters such as the local airway diameter and the angle between the suction channel outlet axis and the airway centerline.

[0046] Based on the reconstructed 3D point cloud model, combined with the segmentation mask of the secretion region in the 2D image, the volume of the secretion target is estimated by integration or voxelization methods. For irregularly shaped secretions, ellipsoid fitting or voxel accumulation methods are used for volume calculation.

[0047] The formula for calculating the airway configuration factor (ACF) is as follows:

[0048] This factor quantitatively characterizes the impact of the current airway geometry on the efficiency of suction transmission: the smaller the airway diameter and the larger the angle between the suction direction and the secretion adhesion surface, the higher the required negative pressure value.

[0049] Adaptive negative pressure target value The calculation formula is:

[0050] In the formula, The baseline negative pressure value is set based on clinical guidelines or operator experience. It is the geometric attenuation coefficient, with a value ranging from 0 to 1, used to adjust the reduction intensity of negative pressure by geometric factors; For reference airway diameter, it can be preset to the typical main bronchus diameter; The local diameter of the airway at the location of the secretion is obtained by combining endoscopic images with monocular ranging or structured light reconstruction methods. The angle between the suction channel outlet axis and the airway centerline reflects the geometric relationship between the suction direction and the secretion adhesion surface; This is the volume adjustment coefficient, typically taken as a value within... arrive Between these, the contribution of secretion volume to the target negative pressure is used to regulate the amount of secretion. The volume estimate of the secretion target is obtained based on image segmentation and 3D reconstruction; For reference, the volume of secretions can be preset to the volume of a typical sputum plug.

[0051] The operation decision module will calculate the results. The negative pressure setting value for this suction operation, along with preset suction duration, waveform, and other parameters, generates a suction command and sends it to the suction execution module. The suction execution module adjusts the working state of the negative pressure source according to the command, ensuring that the negative pressure value within the suction channel reaches and stabilizes at a certain level. Nearby, it exerts an attraction on the target of the secretion.

[0052] During the suction process, the safety monitoring module monitors the actual negative pressure value within the suction channel and the dynamic image characteristics of the secretion target in real time. If the secretion is detected as not being removed within a preset time, the system will re-execute the above steps, update the geometric and volume parameters based on the secretion residue, recalculate the target negative pressure value, and dynamically correct it within the safety threshold range until the secretion is successfully removed or a safety interruption condition is triggered.

[0053] An adaptive negative pressure control method based on airway geometry and fluid dynamics simulation was used to achieve precise matching between suction parameters and local airway anatomical features. Compared with conventional fixed negative pressure suction technology, this method significantly reduces the negative pressure impact on distal airways while ensuring effective secretion clearance. A comparative experiment was conducted using a lung model combined with isolated porcine trachea specimens. The control group underwent fixed negative pressure suction; the experimental group used an adaptive negative pressure calculation formula to dynamically adjust the target negative pressure value according to the local airway diameter and suction direction. Experimental setup: Equal volumes of artificial simulated sputum specimens were placed at three typical locations: the main bronchus, lobar bronchus, and segmental bronchus. The suction success rate and the peak negative pressure experienced by the airway walls during suction were recorded. The experimental results showed that: at the main bronchus, the negative pressure values ​​of the two groups were similar (control group -100 mmHg, experimental group -96 mmHg), and the clearance success rate was 100% for both groups; at the lobar bronchus, the negative pressure of the control group remained at -100 mmHg, with a clearance success rate of 100%, but the peak negative pressure on the airway wall reached -112 mmHg, while the experimental group adaptively adjusted the negative pressure to -82 mmHg, achieving a clearance success rate of 96.7% and reducing the peak negative pressure on the airway wall by 26.8%; at the segmental bronchus, the negative pressure of the control group remained at -100 mmHg, with a clearance success rate of 93.3%, but the peak negative pressure on the airway wall reached as high as -158 mmHg, exceeding the generally accepted safety threshold, while the experimental group adaptively adjusted the negative pressure to -65 mmHg, achieving a clearance success rate of 83.3% and reducing the peak negative pressure on the airway wall by 58.9%. The results indicate that by introducing an airway configuration factor to adaptively adjust the negative pressure, the negative pressure impact on the distal airway can be controlled within a safe range while ensuring an acceptable clearance success rate, effectively reducing the risk of airway mucosal damage caused by excessive local negative pressure.

[0054] Example 4: Building upon the aforementioned embodiments, and addressing the shortcomings of existing methods that rely solely on single image features or simple pressure monitoring, failing to comprehensively assess potential risks during suctioning, a safety control method based on multimodal sensor fusion for coordinated monitoring of airway wall contact force and suction resistance is proposed. This method generates measurable characteristic signals across multiple physical dimensions when abnormal contact occurs between the bronchoscope tip and the airway wall, or when abnormal changes in fluid resistance occur in the suction channel due to secretion blockage. These abnormal states include optical image features, mechanical contact features, and fluid dynamics features. By fusing mucosal deformation features extracted from endoscopic images, contact force vectors detected by fiber optic force sensors, and pressure-flow dynamic response curves within the suction channel, a three-dimensional safety boundary model is constructed, enabling multi-dimensional safety monitoring of the entire suctioning process. When any monitored parameter in any dimension exceeds a preset safety threshold, the system determines it as an abnormal state and immediately terminates automatic operation. Simultaneously, it outputs a manual intervention prompt containing the abnormality type and location information, providing the operator with accurate fault location and decision support.

[0055] A miniature fiber optic force sensor array is integrated at the tip of the bronchoscope to detect the magnitude and direction of the contact force between the endoscope and the airway wall in real time. Pressure and flow sensors are integrated at the suction channel inlet to monitor the negative pressure and airflow rate during suction in real time. The endoscope image acquisition module retains standard image acquisition functions. All of these sensors establish real-time data communication with the safety monitoring module.

[0056] The security monitoring module has a pre-set three-dimensional security boundary model, which is composed of threshold ranges in the following three dimensions: Optical dimension threshold: Based on the mucosal deformation degree score, mucosal congestion index and secretion recognition confidence level output by the image recognition module, the threshold for judging optical abnormalities is set. Mechanical dimension thresholds are set based on the amplitude and direction of the contact force detected by the fiber optic force sensor, including the maximum allowable contact force threshold, the continuous contact time threshold, and the impact force change rate threshold. Fluid dimension thresholds are set based on the negative pressure-flow dynamic response curve within the suction channel to define abnormal suction resistance thresholds, including the pressure build-up time during the suction initiation phase, the flow fluctuation range during the steady-state phase, and the characteristic pressure fluctuation pattern of secretions as they pass through the suction channel.

[0057] Throughout the automated suctioning process, the safety monitoring module controls each sensor to collect data via synchronous triggering. The endoscopic image acquisition module acquires real-time images at a preset frame rate, and the image recognition module extracts feature parameters such as the area of ​​mucosal deformation and mucosal color changes from these images. The fiber optic force sensor array acquires contact force data at a preset sampling frequency, and after filtering, extracts the force amplitude, force direction angle, and rate of change of force. The pressure sensor and flow sensor synchronously acquire pressure-flow data within the suction channel, and after feature extraction, obtain parameters such as pressure settling time, steady-state pressure fluctuation range, and pressure-flow correlation coefficient.

[0058] The safety monitoring module compares the extracted feature parameters with the corresponding thresholds in the 3D safety boundary model and executes the following anomaly detection logic: If the mucosal deformation score exceeds the optical threshold, or the confidence level of secretion recognition drops sharply to below the preset lower limit, it is judged as an optical dimension abnormality; If the amplitude of the contact force exceeds the mechanical threshold, or if the direction of the contact force is directed toward the airway wall and the contact time exceeds the preset duration, it is determined to be an abnormality in the mechanical dimension. If the pressure build-up time after suction starts exceeds the preset range, or the steady-state pressure fluctuation exceeds the limit, or the pressure-flow curve shows a specific blockage pattern, it is determined to be an abnormality in the fluid dimension.

[0059] When any dimension is determined to be abnormal, the security monitoring module further integrates abnormal information from multiple dimensions to perform status assessment and classification: Mild anomaly: A single dimension briefly exceeds the limit and can recover on its own. The system issues a prompt message but does not interrupt automatic operation. Moderate abnormality: If a single dimension continues to exceed the limit or two dimensions show mild abnormalities at the same time, the system will immediately stop automatic operation, maintain the current bronchoscope position, and output a prompt message containing the abnormal dimension and suggested treatment methods. Severe abnormality: If two or more dimensions show obvious abnormalities at the same time, or if any dimension shows an emergency abnormality that may cause tissue damage, the system will immediately stop automatic operation and automatically execute the bronchoscopy withdrawal procedure, withdrawing the bronchoscope to the preset safe position, and at the same time issuing an emergency manual intervention request.

[0060] After issuing a prompt for manual intervention, the system switches control to manual operation mode, while the safety monitoring module continues to monitor subsequent operations. Simultaneously, the system automatically records the time, location, monitoring data across various dimensions, and the specific parameters that triggered the anomaly, storing these in the operation log for later analysis or system optimization.

[0061] A safety control method based on multimodal sensor fusion for coordinated monitoring of airway wall contact force and suction resistance was implemented, achieving multi-dimensional and highly sensitive safety monitoring of the entire suctioning process. Compared with relying solely on single pressure monitoring or simple image recognition, improvements were made in the accuracy of abnormal state identification and the timeliness of manual intervention. The comparative experiment was conducted as follows: In 30 patients undergoing bronchoscopic suctioning, three senior respiratory therapists observed and recorded abnormal events in real time, serving as the standard. Simultaneously, the system ran the conventional safety monitoring module and the multimodal sensor fusion unit in parallel, comparing their ability to identify abnormal events. A total of 87 abnormal events were recorded during the experiment, including: excessively deep airway wall contact (contact force > 0.5 N for more than 2 seconds) in 32 cases, suction channel obstruction (pressure build-up time > 1.5 seconds or abnormal flow fluctuations) in 28 cases, and a sudden drop in confidence in secretion identification (due to blood or mucus obstructing the lens) in 27 cases. The identification results showed that the conventional pressure monitoring module could only identify suction channel blockage abnormalities, and could not identify the other two types of abnormalities at all, with an overall abnormality identification rate of 32.2%. The multimodal sensor fusion unit achieved an accuracy rate of 93.8% in identifying airway wall contact abnormalities, 96.4% in identifying suction channel blockage abnormalities, and 88.9% in identifying image recognition confidence abnormalities, with an overall abnormality identification rate of 93.1%, which is 189.1% higher than the conventional method. In terms of the timeliness of manual intervention, the average response time of the multimodal sensor fusion unit from the occurrence of an abnormality to the issuance of an intervention prompt was 0.8 ± 0.3 seconds, which is 83.0% shorter than the delay time of manual observation in detecting abnormalities. These results indicate that by fusing optical, mechanical, and fluid dynamic information for collaborative monitoring, accurate identification and immediate response to various operational abnormalities can be achieved, significantly improving the safety level of automated suctioning operations and effectively reducing the risk of complications such as airway damage and hypoxemia caused by persistent abnormal conditions.

Claims

1. An automated suction bronchoscopic system for clearing airway secretions, characterized in that, include: The endoscopic image acquisition module is used to acquire real-time images of the airway lumen distal to the bronchoscope. The secretion recognition module is used to analyze the images inside the airway cavity to identify secretion targets and output their aspirability characteristic parameters; The operation decision module is used to determine whether the current secretion target is within the preset airway level range where suction operation can be performed based on the suctionability characteristic parameters, and to generate a suction command when the determination is affirmative. A suction execution module is used to apply negative pressure to the target secretion through the suction channel of the bronchoscope in response to the suction command; The safety monitoring module is used to monitor the status parameters of the attraction process in real time, and to interrupt the automatic operation and issue a manual takeover request when an abnormal state is detected. The operation path recording module is used to identify bronchial bifurcation markers based on a continuous sequence of endoscopic images, construct the topology of the current operation path, and mark the bronchial branches for which suction operations have been performed.

2. The system according to claim 1, characterized in that: The system is suitable for patients who require long-term mechanical airway clearance due to weakened or absent cough reflexes.

3. The system according to claim 1, characterized in that: The secretion recognition module distinguishes mobile sputum secretions from fixed airway tissues or foreign bodies based on at least one of the morphological features, light reflection characteristics, or fluidity characteristics of secretions in optical images.

4. The system according to claim 3, characterized in that: After applying a preset weak airflow disturbance, the secretion recognition module generates a quantitative attractiveness assessment index based on at least one of the morphological features, light reflection characteristics, or fluidity characteristics of the secretion in the optical image, and classifies the secretion target into different attractiveness levels accordingly, so that the operation decision module can execute a graded attraction strategy.

5. The system according to claim 1, characterized in that: The preset airway hierarchy range for suction operations includes the main bronchus, lobar bronchus, and segmental bronchus. The operation decision module identifies the current hierarchy based on the anatomical morphological characteristics of bronchial bifurcation and prohibits automatic suctioning of terminal bronchus outside this range.

6. The system according to claim 1, characterized in that: The suction execution module adaptively adjusts the intensity parameters of negative pressure suction based on the local geometry of the airway where the secretion is located, including the airway diameter and the angle between the suction direction and the airway centerline, combined with the spatial volume characteristics of the secretion target.

7. The system according to claim 1, characterized in that: The safety monitoring module includes a multimodal sensor fusion unit, which is used to simultaneously acquire endoscopic image features, contact force features between the bronchoscope tip and the airway wall, and fluid dynamic features within the suction channel. The safety monitoring module identifies abnormal states during the suctioning process based on the collaborative analysis of multimodal features, and executes graded interruption responses according to the overall severity of the abnormality.

8. The system according to claim 7, characterized in that: The abnormal states include: the secretion target not being cleared after a preset number of suction operations; the confidence level of the secretion recognition module in recognizing the target in the current field of view being lower than a threshold; or the negative pressure response curve detected by the suction execution module deviating from the preset fluid dynamic characteristics.

9. The system according to claim 1, characterized in that: After issuing a manual takeover request, the security monitoring module switches system control to manual operation mode and continues to monitor the status of subsequent operations until manual confirmation is received to restore automatic mode.

10. An automated bronchoscope method for clearing airway secretions, characterized in that: The method is implemented based on the system described in any one of claims 1-9: The method includes: Real-time acquisition of images of the airway lumen distal to the bronchoscope; The images inside the airway are analyzed to identify the target secretion and output its aspirability characteristics. Based on the aspirability characteristic parameters, determine whether the current secretion target is within the preset airway level range where aspiration can be performed; When the determination is positive, a suctioning instruction is generated and negative pressure is applied to the target secretion through the suction channel of the bronchoscope; The system monitors status parameters in real time during the attraction process. When an abnormal state is detected, the automatic operation is interrupted and a manual takeover request is issued.

Images