Dual camera visual flushing double-lumen bronchial intubation device and collaborative control method

Abstract

The application discloses a double-camera visual flushing double-lumen bronchial intubation device and a collaborative control method, belongs to the technical field of medical devices, and aims to solve the problems of single data acquisition, lagging abnormal identification and easy distraction of doctors by double video streams of the existing device. The method comprises the following steps: capturing the positioning depth of the tube core through the positioner, triggering the double-camera to collect the cuff and lumen video streams and to construct a collaborative data set, triggering the lumen video stream sampling analysis through the tube core stable state judgment, regulating the sampling interval and screening the redundant frames to construct a key frame set, identifying the abnormal frames through the benchmark correction and multi-feature difference calculation, and adaptively switching the display terminal video source based on the abnormal degree. The application realizes accurate identification and efficient monitoring of the airway, takes into account the clinical operation concentration and response timeliness, and significantly improves the safety and operation efficiency of the intubation operation.

Description

Technical Field

[0001] This invention relates to the field of medical device technology, and more specifically to a dual-camera visual flushing dual-lumen endotracheal intubation device and its collaborative control method. Background Technology

[0002] Double-lumen endotracheal intubation is a key medical device in thoracic surgery and anesthesia management of critically ill patients. Its core functions include lung isolation, lateral ventilation, and airway secretion suction. Its effectiveness directly impacts surgical safety and patient ventilation stability. However, with increasing clinical demands for precision medicine, existing double-lumen endotracheal intubation technologies are gradually revealing several limitations:

[0003] Existing devices often rely on single-dimensional data monitoring, making it difficult to simultaneously and comprehensively capture crucial information such as cuff fit, bronchial cavity structure, and lung mucosal morphology. This results in physicians being unable to establish a complete understanding of the overall airway condition, increasing the uncertainty of operational decisions. Traditional methods often rely on physicians manually observing video streams, which is not only labor-intensive but also susceptible to interference from surgical procedures and visual fatigue. This makes it difficult to quickly and accurately identify potential risks such as airway mucosal congestion, secretion accumulation, and abnormal lung morphology, often leading to missed diagnoses or delayed interventions. Regarding display and analysis coordination, even devices with dual-channel video acquisition capabilities lack scientific display and analysis triggering logic. Either simultaneous display of two video streams distracts the physician, affecting focus on core operations such as intubation positioning and cuff inflation; or single-channel display results in missing information. Furthermore, the timing of analysis triggers lacks precise judgment criteria, initiating analysis before the intubation position is stable, reducing the reliability of anomaly detection and causing unnecessary waste of computational resources. These problems make it difficult for existing dual-lumen endotracheal intubation devices to simultaneously achieve high levels of operator focus, comprehensive information coverage, and timely response to abnormalities in clinical applications, thus limiting their effectiveness in precision medicine scenarios. Therefore, to overcome these limitations, this invention proposes a dual-camera-guided visual flushing dual-lumen endotracheal intubation device and a collaborative control method. Summary of the Invention

[0004] To address the shortcomings of existing technologies, the present invention aims to provide a dual-camera visual flushing dual-lumen endotracheal intubation device and a collaborative control method, which solves the problem of how to comprehensively acquire airway and cuff-related status information and accurately identify airway abnormalities during dual-lumen endotracheal intubation operations, while avoiding interference with the doctor's focus and balancing the stability of clinical operations with the timeliness of abnormal response.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] A collaborative control method for a dual-camera visual flushing dual-lumen endotracheal intubation device includes:

[0007] The positioning depth of the stent within the cannula is captured in real time to determine whether real-time video data acquisition is triggered. If so, video stream data of the cuff and video stream data of the cavity are collected separately to construct a collaborative dataset.

[0008] Calculate the change in the core positioning depth within the preset monitoring window, identify stable monitoring windows, and use this information to determine whether to trigger the cavity video stream sampling analysis operation.

[0009] When the cavity video stream sampling analysis operation is triggered, the key frame images are extracted from the cavity video stream data of the collaborative dataset by adjusting the sampling interval, and the set of key frame images is constructed by combining validity screening and redundancy removal optimization.

[0010] The system locates the baseline keyframe in the keyframe image set, extracts stable feature points from the baseline keyframe, and establishes a feature coordinate system for baseline correction of newly added keyframes in the keyframe image set. By calculating the feature difference index between the corrected new keyframes and the baseline keyframes, potential abnormal frames are identified. Then, the system performs abnormal region segmentation and quantization on the potential abnormal frames to determine the abnormal keyframes.

[0011] A dual-video-stream adaptive display mechanism is constructed, which adaptively switches the video source of the display terminal based on the degree of abnormality of abnormal keyframes and a collaborative dataset.

[0012] Specifically, the steps for constructing a collaborative dataset include:

[0013] When the dual-camera visual flushing dual-lumen endotracheal intubation device is started, the intelligent control module triggers the data acquisition process and activates the data acquisition components.

[0014] The system captures the positioning depth of the die within the cannula in real time. When the positioning depth of the die exceeds a preset depth threshold, real-time video data acquisition is triggered.

[0015] The intelligent control module sends a start command to the dual cameras, which include a main camera and a secondary camera; the main camera is used to collect video stream data of the endoscopic cuff; the secondary camera is used to collect video stream data of the cavity.

[0016] The acquisition timestamps of the endoscopic cuff video stream data and the cavity video stream data are extracted separately. The timestamps of the endoscopic cuff video stream data and the cavity video stream data are calibrated and aligned using the internal clock of the intelligent control module as a unified time reference.

[0017] Based on the chip positioning depth, the chip positioning depth data is bound to the capsular video stream data and cavity video stream data according to the acquisition timestamp to form a collaborative dataset.

[0018] Specifically, the steps for determining whether to trigger the cavity video stream sampling analysis operation include:

[0019] The chip positioning depth is received in real time, and the chip positioning depth is sampled at preset time intervals to construct a chip positioning depth sequence;

[0020] Simultaneously, a monitoring window is configured to filter and denoise the sampled die positioning depth sequence within the monitoring window, extract the extreme values ​​of the filtered and denoised die positioning depth sequence, and calculate the extreme value difference as the variation range of die positioning depth within the monitoring window.

[0021] If the change is less than the preset fluctuation threshold, the monitoring window is marked as a stable monitoring window; the cumulative duration of the stable monitoring window and the interval between two adjacent stable monitoring windows are calculated.

[0022] If the cumulative duration of the stable monitoring window is greater than the minimum stable cumulative duration, and the interval duration of all adjacent stable monitoring windows is less than or equal to the maximum allowable interval duration, then the die positioning depth is determined to be in a stable state, triggering the cavity video stream sampling analysis operation.

[0023] Specifically, the steps for adjusting the sampling interval include:

[0024] Set an initial sampling interval. When the cavity video stream sampling analysis operation is triggered, extract key frames frame by frame from the cavity video stream data in the collaborative dataset according to the initial sampling interval to form an initial key frame candidate set.

[0025] A difference analysis was performed on consecutive keyframe images in the initial keyframe candidate set. The gray-level difference, structural similarity index and feature point matching degree of the corresponding pixel regions of consecutive keyframe images were calculated respectively, and the inter-frame difference value was generated by normalization.

[0026] The inter-frame difference value is compared with a preset difference threshold, which includes a preset lower difference threshold and a preset upper difference threshold. If the difference value is less than the preset lower difference threshold, a sampling reduction magnitude is generated based on the degree of deviation between the difference value and the preset lower difference threshold to increase the sampling interval.

[0027] If the difference value is greater than the preset difference upper threshold, a sampling increase is generated based on the degree of deviation between the difference value and the preset difference upper threshold, which is used to shorten the sampling interval.

[0028] Specifically, the steps for constructing the keyframe image set include:

[0029] The initial keyframe candidate set is filtered for effectiveness. The mean grayscale gradient, edge contour sharpness and signal-to-noise ratio of the key frames in the initial keyframe candidate set are calculated and compared with the corresponding preset standard thresholds to identify and remove blurry frames.

[0030] The initial keyframe candidate set after validity screening is optimized by redundancy removal. Image similarity is calculated for consecutive keyframes. If the image similarity of consecutive keyframes is higher than the preset redundancy threshold, the previous keyframe is retained as a valid keyframe and the subsequent redundant keyframe is removed; otherwise, all are retained as valid keyframes to form a keyframe image set.

[0031] Specifically, the steps for benchmark correction of newly added keyframes in the keyframe image set include:

[0032] Locate the reference keyframe of the keyframe image set and perform image preprocessing on the reference keyframe;

[0033] For the preprocessed reference keyframe, extract stable feature points of the reference keyframe;

[0034] Based on the local pixel grayscale distribution and gradient direction of stable feature points, a binary feature vector is generated for each stable feature point.

[0035] Using any stable feature point as the core reference point, calculate the straight-line distance and relative direction of the other stable feature points relative to the core reference point to determine the relative positional relationship of each stable feature point; and establish a feature coordinate system with the core reference point as the origin.

[0036] For newly added keyframes sampled after the baseline keyframe in the keyframe image set, image preprocessing is performed, and new feature points of the new keyframes are extracted, and binary feature vectors of each new feature point are generated.

[0037] Based on the feature vector of the newly added feature point and the binary feature vector of each stable feature point in the baseline keyframe, the feature point similarity is calculated. If the feature point similarity between the newly added feature point and the stable feature point is greater than the similarity threshold, the corresponding newly added feature point and the stable feature point are determined to be a matching feature point pair.

[0038] Based on the feature coordinate system, spatial coordinate mapping and deviation calculation are performed on the matching feature point pairs. By solving the homography matrix, correction parameters for the newly added keyframes are generated, including scaling parameters, translation parameters, and rotation parameters. The newly added keyframes are then corrected based on the correction parameters.

[0039] Specifically, the steps for identifying potentially anomalous frames include:

[0040] The feature difference index is calculated between the corrected new keyframe and the baseline keyframe. The feature difference index includes: color feature difference index, texture feature difference index and contour feature difference index.

[0041] The color feature difference index is obtained by calculating the color deviation value of the HSV color space parameters between the corrected new keyframe and the reference keyframe.

[0042] The texture feature difference index is obtained by calculating the absolute difference between the texture parameters of the cavity wall and mucosal surface in the corrected new keyframe and the reference keyframe.

[0043] The contour feature difference index is obtained by extracting the contour parameters of the edge contours of the corrected new keyframe and the reference keyframe, and normalizing the absolute deviation of each contour parameter.

[0044] Each feature difference index has a preset normal fluctuation threshold. If the deviation value of any feature difference index of a newly added key frame is greater than the corresponding normal fluctuation threshold, the newly added key frame is marked as a potential abnormal frame, and the feature deviation type that exceeds the normal fluctuation threshold is recorded.

[0045] Specifically, the steps for segmenting and quantizing potential anomalous frames to determine anomalous keyframes include:

[0046] Based on the type of feature deviation, feature images of the corrected new keyframe and the reference keyframe are generated respectively; the feature images of the new keyframe and the reference keyframe are differentially processed to generate a difference image, and the effective threshold of each pixel in the difference image is determined by an adaptive threshold segmentation algorithm.

[0047] Pixels with gray values ​​greater than the effective threshold in the difference image are marked as difference pixels. Then, the discrete difference pixels are aggregated into continuous abnormal regions by the region growing algorithm. At the same time, false regions with areas smaller than the preset minimum region area are removed to obtain the abnormal regions of potential abnormal frames.

[0048] Calculate the relative area ratio of the abnormal region in the potential abnormal frame. If it is greater than the preset area threshold, the potential abnormal frame is determined to be an abnormal key frame. Then, perform dynamic change analysis on the abnormal key frame, quantify the abnormal change trend, and determine the abnormal state.

[0049] Specifically, the steps for adaptively switching the video source on the display terminal include:

[0050] The abnormal keyframes identified are classified according to their degree of abnormality, and based on the degree of abnormality, video source switching rules are set, switching instructions are generated, and sent to the display terminal.

[0051] After receiving the switching command, the display terminal pauses the playback of the default video stream in the display area, switches the video source of the display area from the endoscopic video stream channel to the cavity video stream channel, and automatically locates the abnormal key frame in the collaborative data set to overlay the abnormal area magnified image of the abnormal key frame.

[0052] When a manual switch is triggered, the display terminal receives the manual switch signal and interrupts the current adaptive switch process: if the current display area is the endoscopic video stream, it switches to the endoscopic video stream and overlays an enlarged view of the abnormal area of ​​the abnormal keyframe; if the current display area is the endoscopic video stream, it switches back to the endoscopic video stream.

[0053] The dual-camera visual flushing dual-lumen endotracheal intubation device includes a data acquisition module, an intelligent control module, a keyframe processing module, an anomaly detection module, and a display linkage module.

[0054] The data acquisition module is used to acquire real-time video stream data from the endoscopic tube, cavity, and core positioning depth. The intelligent control module is used to construct a collaborative dataset, determine whether to trigger cavity video stream sampling analysis, and build a dual-video stream adaptive display mechanism. The keyframe processing module is used to extract keyframe images from the cavity video stream data by adjusting the sampling interval and construct a keyframe image set. The anomaly detection module is used to perform baseline correction on newly added keyframes in the keyframe image set. By calculating the feature difference index between the corrected new keyframes and the baseline keyframes in the keyframe image set, potential abnormal frames are identified. The module also performs abnormal region segmentation and quantization on potential abnormal frames to determine abnormal keyframes. The display linkage module is used to adaptively switch the video source of the display terminal.

[0055] The beneficial effects of this invention are:

[0056] This application precisely triggers dual video stream acquisition by positioning the core tube at a depth and constructs a collaborative dataset with associated temporal and location information. Based on a stable monitoring window, it triggers cavity video stream sampling analysis. Combined with dynamic sampling interval adjustment, effectiveness screening, and redundancy removal optimization, it constructs a high-quality keyframe set. Abnormal frames are accurately identified and quantified through benchmark correction and multi-dimensional feature difference calculation. Then, based on the degree of abnormality, it achieves adaptive display switching between dual video streams. This effectively solves the problems of single data acquisition, delayed abnormality identification, display interference with operational focus, and inaccurate analysis triggering in existing technologies. It not only comprehensively acquires complete status information of the airway and cuff but also improves the accuracy and timeliness of abnormality identification, while avoiding the waste of ineffective computing power. It balances the focus of clinical operation and surgical safety, and significantly optimizes the clinical application effect of dual-lumen endotracheal intubation. Attached Figure Description

[0057] Figure 1 This is a flowchart of the collaborative control method for the dual-camera visual flushing dual-lumen endotracheal intubation device of the present invention;

[0058] Figure 2 This is a flowchart illustrating the process of determining whether a cavity video stream sampling analysis operation is triggered according to the present invention.

[0059] Figure 3This is a flowchart illustrating the keyframe image extraction process for cavity video stream data according to the present invention.

[0060] Figure 4 This is a flowchart illustrating the baseline correction of newly added keyframes in the keyframe image set according to the present invention.

[0061] Figure 5 This is a flowchart illustrating the process of segmenting and quantizing potentially abnormal frames to determine abnormal keyframes. Detailed Implementation

[0062] Please see Figure 1 This embodiment describes a collaborative control method for a dual-camera visual flushing dual-lumen endotracheal intubation device, including:

[0063] Step S1: After the dual-camera visual flushing dual-lumen endotracheal intubation device is started, the intelligent control module triggers multi-source data acquisition and activates the data acquisition components, including dual cameras, a locator, cuff pressure sensor, airway environment sensor, ventilation flow sensor, and device status detector. The locator captures the stent positioning depth in the intubation tube in real time to determine whether real-time video data acquisition is triggered. If so, the intelligent control module sends a start command to the dual cameras to collect cuff video stream data and cavity video stream data respectively to build a collaborative dataset. The intelligent control module is the core control component that integrates data reception, parsing and processing, logical judgment, and command generation functions. It receives raw data transmitted from each data acquisition component, performs format conversion and feature extraction on the data, judges trigger conditions according to preset logic, issues operation commands to relevant components, and coordinates the collaborative operation of each acquisition component and execution component. The dual cameras include a main camera and a secondary camera. The main camera is positioned near the cuff on the intubation tube. After activation, it focuses on the area surrounding the cuff and continuously collects video stream data of the cuff, including the contour features of the bronchial wall, the initial installation position of the cuff, and the surrounding environment. The secondary camera is positioned at the patient end of the intubated bronchus and faces the inside of the bronchial cavity and the lung region after activation. It continuously collects video stream data of the bronchial cavity wall structure, the morphology of the lung mucosa surface, and ventilation-related cavity data.

[0064] In this embodiment, the dual-camera startup is precisely triggered by the locator, avoiding invalid video data collection. At the same time, multi-dimensional sensor data and dual-channel targeted video streams are integrated to construct a collaborative dataset. This not only solves the problems of single data collection and lack of precise basis for camera startup timing in the existing technology, but also provides comprehensive and synchronous basic data support for subsequent collaborative operations such as tunic positioning and anomaly monitoring, reducing errors caused by manual judgment.

[0065] Preferably, the specific steps for constructing a collaborative dataset include:

[0066] When the dual-camera visual flushing dual-lumen endotracheal intubation device is started, the intelligent control module triggers the data acquisition process, activates the data acquisition components, and after the locator is started, it captures the positioning depth of the stent in the intubation tube in real time through the built-in displacement detection element. The intelligent control module calls the preset judgment logic to compare the received positioning depth data in real time. When the stent positioning depth is greater than the preset depth threshold, it indicates that the tip of the stent has reached the target area in the bronchus where effective observation and data acquisition can be achieved, triggering real-time video data acquisition. The intelligent control module sends a start command to the dual cameras. The main camera focuses on the area around the cuff to collect cuff video stream data, and the secondary camera faces the inside of the bronchial cavity and the lung area to collect cavity video stream data. The intelligent control module caches the two types of raw video data into independent preset cache partitions according to the video stream type. Otherwise, the dual cameras remain in standby mode, and the locator continues to monitor and transmit stent positioning depth data. Among them, the preset depth threshold refers to the minimum relative position parameter of the stent to the intubation tube body when the two cameras can respectively obtain clear images of the cuff periphery and the bronchial cavity. It is preset based on the device structural design parameters, bronchial anatomical characteristics and clinical application requirements.

[0067] The intelligent control module extracts the acquisition timestamps from both the endoscopic cuff video stream and the intracavitary cavity video stream. Using the internal clock of the intelligent control module as a unified time reference, it calibrates and aligns the timestamps of the endoscopic cuff video stream and the intracavitary cavity video stream. This ensures the correlation and consistency of the two video data streams in the temporal dimension, providing an accurate time-series data foundation for subsequent collaborative analysis.

[0068] Based on the sclereographic positioning depth, the intelligent control module constructs a correlation mapping system between the sclereographic video stream data and the cavity video stream data. It binds the sclereographic positioning depth data with the sclereographic video stream data and the cavity video stream data according to the acquisition timestamp, and clarifies the acquisition position and observation scene corresponding to the two types of video stream data at different sclereographic positioning depths. This enables the two video stream data to form a mutually corroborating and scene-clearly defined correlation data set, providing data support for subsequent collaborative judgment based on acquisition position and visual state.

[0069] The intelligent control module performs structured storage of the cuff video stream data and cavity video stream data after time calibration and correlation mapping. According to the hierarchical structure of acquisition time period, chip positioning depth range, and video data type, the cuff video stream data and cavity video stream data are stored in the corresponding partitions of the preset database. The acquisition time period is divided according to the time sequence of data reception, the chip positioning depth range is divided according to the preset depth threshold and the actual monitored depth variation range, and the video data type is clearly identified as cuff video stream and cavity video stream to ensure the orderly and traceable storage of data.

[0070] The intelligent control module automatically generates a collaborative dataset index. This index includes the identifier for each video stream segment, the corresponding video data type, the associated chip positioning depth range, the acquisition time span, the database storage path, and data integrity verification information. The index is stored synchronously with the collaborative dataset. The intelligent control module can quickly locate and retrieve the target video stream data through the index, avoiding retrieval delays caused by large data volumes and ensuring efficient data retrieval during subsequent collaborative control processes, thus completing the construction of the collaborative dataset.

[0071] Step S2: The intelligent control module continuously monitors the stent positioning depth data transmitted by the locator. By calculating the change range of the stent positioning depth within the preset monitoring window, it determines whether to trigger the cavity video stream sampling analysis operation. When the change range of the stent positioning depth meets the preset stability condition, it indicates that the cuff has started inflating and the insertion position is fixed, and the stent is in a stable state. The intelligent control module triggers the cavity video stream sampling analysis operation and simultaneously starts the dual video stream adaptive display control logic. If the stent positioning depth is still fluctuating, the initial acquisition state of the dual video streams is maintained, the cavity video stream analysis is not started temporarily, and the cuff video stream continues to be transmitted to the display terminal for the doctor to observe in real time.

[0072] In this embodiment, the cavity video stream sampling analysis is triggered by the stable state of the core positioning depth, ensuring that effective analysis is only carried out after the cuff inflation is completed and the cannula position is fixed, thus avoiding unnecessary computational waste. At the same time, an adaptive display mechanism is adopted, which defaults to displaying the cuff video stream and performs intelligent background analysis of the cavity video stream. This not only solves the problem of the two video streams being displayed separately, which distracts doctors and wastes medical resources, but also achieves accurate early warning of cavity abnormalities through frame sampling analysis and abnormal trigger switching. This allows doctors to focus on core operations without overlooking potential risks, thus balancing the focus and safety of clinical operations.

[0073] Please see Figure 2 Preferably, the specific steps for determining whether to trigger the cavity video stream sampling analysis operation include:

[0074] The system receives the positioning depth of the ductus arteriosus in real time, samples the positioning depth at preset time intervals to construct a positioning depth sequence, and configures a monitoring window. The duration of the monitoring window is set based on the typical time it takes for the ductus arteriosus to stabilize during clinical cannulation operations, ensuring that it can cover the complete process from fluctuation to stability of the ductus arteriosus. Each monitoring window contains at least a preset number of ductus arteriosus positioning depth sampling data, providing a sufficient data foundation for subsequent calculation of the change amplitude.

[0075] After each monitoring window ends, the intelligent control module processes the sampled chip positioning depth sequence within the monitoring window. It uses a sliding filter algorithm to filter and denoise the chip positioning depth sequence, eliminating instantaneous interference signals and avoiding judgment deviations caused by signal fluctuations.

[0076] The extreme values ​​of the filtered and denoised stent positioning depth sequence are extracted, and the difference between the extreme values ​​is calculated as the variation range of the stent positioning depth within the monitoring window. If the variation range is less than the preset fluctuation threshold, it indicates that the stent positioning depth within the monitoring window is in a state of small fluctuation, which meets the preliminary condition of fixed cannulation position, and the monitoring window is marked as a stable monitoring window. The preset fluctuation threshold is pre-set and stored in the intelligent control module based on the device cannulation positioning accuracy requirements and clinical operation specifications.

[0077] The system calculates the cumulative duration of stable monitoring windows and the interval between two adjacent stable monitoring windows. The cumulative duration of stable monitoring windows is the sum of the durations of all windows marked as stable monitoring windows, and the interval is the time difference between the end of the previous stable monitoring window and the start of the next stable monitoring window. Simultaneously, it calls preset stability judgment parameters, which include a minimum cumulative stable duration and a maximum allowable interval duration. The minimum cumulative stable duration is set based on the clinically necessary duration for the stylet to remain stable after cuff inflation, and the maximum allowable interval duration is set based on the critical duration for continued stability after a brief fluctuation in the stylet.

[0078] The intelligent control module compares the cumulative duration of the stable monitoring window with the minimum stable cumulative duration, and also compares the interval duration between adjacent stable monitoring windows with the maximum allowable interval duration. If the cumulative duration of the stable monitoring window is greater than the minimum stable cumulative duration, and the interval duration of all adjacent stable monitoring windows is less than or equal to the maximum allowable interval duration, then the die positioning depth is determined to be continuously stable, meeting the trigger conditions for cavity video stream sampling analysis. The intelligent control module immediately generates a cavity video stream sampling analysis operation trigger command.

[0079] Step S3: After triggering the cavity video stream sampling analysis, based on the previously constructed collaborative dataset, the sampling interval is dynamically adjusted according to the inter-frame differences of the cavity video stream. For slow-changing anomalies, such as mucosal congestion and slow accumulation of secretions, the interval is appropriately increased to reduce redundant frames. For fast-changing anomalies, the interval is shortened to ensure coverage of the abnormal process. Keyframes are initially extracted from the cavity video stream to form a candidate set. Subsequently, through the effectiveness screening step, blurred frames caused by lens contamination and secretion obstruction are removed based on criteria such as image clarity and integrity of the core observation area. Then, through the redundancy removal step, duplicate frames with no substantial difference in airway status are removed based on image similarity. Finally, a keyframe image set with a concise structure, complete information, and meeting quality standards is constructed, providing reliable basic image data support for subsequent accurate anomaly judgment of the cavity video stream.

[0080] In this embodiment, by dynamically adjusting the sampling interval, both the waste of computing power by redundant frames in slow-changing abnormal scenarios and the coverage of key frames in fast-changing abnormal scenarios are avoided. Combined with validity screening and redundancy removal, the image quality and information uniqueness of key frames are further guaranteed. While reducing the computing power consumption of subsequent image analysis, it provides efficient and accurate image data support for subsequent anomaly judgment, taking into account both analysis efficiency and data reliability.

[0081] Please see Figure 3 Preferably, the specific steps for extracting keyframe images from cavity video stream data include:

[0082] Based on the clinical abnormality identification needs and the frame rate parameters of the cavity video stream, the intelligent control module pre-sets the initial sampling interval, which includes a time interval and a frame interval. The time interval is set for slow-change abnormalities such as mucosal congestion and slow accumulation of secretions, while the frame interval is set for fast-change abnormalities such as foreign body intrusion into the cavity and sudden changes in lung morphology. The initial sampling interval can be adjusted for clinical scenarios through the parameter configuration module of the intelligent control module.

[0083] When the cavity video stream sampling analysis operation is triggered, the intelligent control module extracts key frames frame by frame from the cavity video stream data in the collaborative dataset according to the preset initial sampling interval, forming an initial key frame candidate set. At the same time, it records the acquisition timestamp, corresponding chip positioning depth data and sampling interval identifier of each initial candidate key frame to ensure the correlation and traceability between the key frames and the collaborative dataset.

[0084] A difference analysis is performed on consecutive keyframe images in the initial keyframe candidate set. The gray-level difference, structural similarity index and feature point matching degree of corresponding pixel regions in consecutive keyframe images are calculated respectively. The difference degree of consecutive keyframe images is comprehensively quantified. The comprehensive quantization result is then normalized to generate an inter-frame difference degree value that can intuitively reflect the degree of difference.

[0085] The inter-frame difference value is compared with a preset difference threshold, which includes a preset lower difference threshold and a preset upper difference threshold. If the difference value is less than the preset lower difference threshold, it indicates that the airway status corresponding to the consecutive key frame images has no significant change. Based on the degree of deviation between the difference value and the preset lower difference threshold, a sampling reduction amplitude is generated to increase the sampling interval. In time interval mode, the interval duration is extended, and in frame interval mode, the number of interval frames is increased to reduce redundant frame extraction.

[0086] If the difference value exceeds a preset upper threshold, it indicates that there may be dynamic changes or potential abnormalities in the airway status. Based on the degree of deviation between the difference value and the preset upper threshold, a sampling increase is generated to shorten the sampling interval. In time interval mode, the interval duration is shortened; in frame interval mode, the number of interval frames is reduced, and the keyframe extraction density is increased to ensure that the complete process of abnormal changes is captured. The preset upper threshold is greater than the preset lower threshold, which is set based on the normal physiological dynamic range of the airway and the minimum clinical abnormality identification threshold, respectively, and stored in the algorithm parameter library of the intelligent control module.

[0087] If the difference value is less than or equal to the preset upper difference threshold and greater than or equal to the preset lower difference threshold, then the initial sampling interval is maintained.

[0088] The intelligent control module performs validity screening on the initial keyframe candidate set extracted after interval adjustment. It calculates the mean grayscale gradient, edge contour sharpness, and signal-to-noise ratio of the keyframes in the initial keyframe candidate set and compares them with the corresponding preset standard thresholds. If any result fails to meet the corresponding preset standard, the frame is determined to be a blurry frame and is removed from the initial keyframe candidate set. Only keyframes that meet all the standards are retained.

[0089] The initial keyframe candidate set after validity screening undergoes redundancy elimination optimization. An image similarity comparison algorithm is invoked to calculate the image similarity of consecutive keyframes. If the image similarity of consecutive keyframes is higher than a preset redundancy threshold, it indicates that the airway states corresponding to the two frames are highly consistent; only the preceding keyframe is retained as a valid keyframe, and the subsequent redundant keyframe is eliminated. If the image similarity is less than or equal to the preset redundancy threshold, all are retained as valid keyframes. Through redundancy elimination, the number of keyframes is further reduced while ensuring anomaly identification coverage, thus decreasing the computational consumption of subsequent image analysis. The final result is a concise, complete, and redundant keyframe image set, providing efficient and accurate image data support for subsequent anomaly judgment in the cavity video stream. The preset redundancy threshold is set based on the judgment requirement that the airway states corresponding to two images are substantially different.

[0090] Step S4: Locate and preprocess the reference keyframe from the keyframe image set, extract its stable feature points to establish a feature coordinate system, perform reference correction on newly added keyframes in the keyframe image set to eliminate physiological motion interference; identify potential abnormal frames by calculating three types of feature difference indicators: color, texture, and contour; generate feature images according to feature deviation type and perform differential analysis, segment and quantize the abnormal regions of potential abnormal frames, determine abnormal keyframes, and finally perform dynamic change analysis and quantification trend analysis on abnormal keyframes to provide accurate abnormal keyframes and dynamic development information for cavity video stream anomaly judgment.

[0091] In this embodiment, benchmark correction effectively eliminates the interference of physiological movements such as breathing and heartbeat on image comparison, ensuring that anomaly analysis focuses only on pathological changes in the airways. Multi-dimensional feature difference calculation and targeted abnormal region segmentation enable accurate identification of conditions such as mucosal congestion, secretion accumulation, and abnormal lung morphology, avoiding the limitations of single feature judgment. Anomaly quantification and dynamic trend analysis provide doctors with clear information on the degree and development status of abnormalities, reducing misjudgments or delays caused by static judgments and providing a scientific basis for whether and how to intervene, significantly improving the reliability of cavity abnormality judgment and the effectiveness of clinical decision-making.

[0092] Please see Figure 4 Preferably, the specific steps for benchmark correction of newly added keyframes in the keyframe image set include:

[0093] The baseline keyframe of the keyframe image set is located and image preprocessing is performed on the baseline keyframe. This includes using a Gaussian filtering algorithm to eliminate high-frequency noise caused by device signal interference and airway secretion reflection, using a histogram equalization algorithm to optimize the color contrast of the bronchial cavity wall and lung mucosa region, and using a bilateral filtering algorithm to perform edge preservation and smoothing. This ensures that the baseline keyframe after image preprocessing not only removes invalid interference but also completely preserves the structural and feature information of the airway core region.

[0094] For the baseline keyframe after image preprocessing, stable feature points of the baseline keyframe are extracted by feature detection algorithm, including fixed corner points of the bronchial cavity wall, feature points of the mucosa and cavity boundary, and feature anchor points of the lung edge. Stable feature points are less affected by physiological movements such as breathing and heartbeat, and have spatial uniqueness in the airway structure.

[0095] Based on the local pixel grayscale distribution and gradient direction of stable feature points, a binary feature vector is generated for each stable feature point. Using any one stable feature point as the core reference point, the straight-line distance and relative direction of the remaining stable feature points relative to the core reference point are calculated to determine the relative positional relationship of each stable feature point. A feature coordinate system is established with the core reference point as the origin. This coordinate system is constructed based on the relative positional relationship of the stable feature points. For example, the direction pointing from the origin to the lung edge feature anchor point is the positive X-axis, and the direction perpendicular to the X-axis and pointing towards the inner side of the mucosa is the positive Y-axis. This does not rely on absolute spatial coordinates and is only used for referencing the feature point positions of unified baseline keyframes and newly added keyframes.

[0096] For newly added keyframes sampled after the baseline keyframe in the keyframe image set, image preprocessing is performed according to the same preprocessing procedure as the baseline keyframe to ensure that the image quality and feature extraction conditions of the two types of keyframes are consistent; and new feature points of the newly added keyframes are extracted and binary feature vectors of each new feature point are generated.

[0097] Based on the feature vectors of newly added feature points and the binary feature vectors of each stable feature point in the baseline keyframe, the similarity of feature points is calculated. For example, Hamming distance is used to measure the similarity of binary feature vectors. The smaller the distance, the higher the feature similarity. A similarity threshold is preset based on the clinical airway feature matching accuracy requirements. If the feature point similarity between a newly added feature point and a stable feature point is greater than the similarity threshold, it is determined to be a matching feature point. All matching feature point pairs that meet the conditions are identified and recorded to provide a positional correspondence basis for the baseline correction of subsequent newly added keyframes.

[0098] Based on the feature coordinate system, spatial coordinate mapping and deviation calculation are performed on the matching feature point pairs. By solving the homography matrix, correction parameters for the new keyframe are generated, including scaling, translation, and rotation parameters. Specifically, the process is as follows: The coordinates of the matching feature points of the reference keyframe are substituted into the preset feature coordinate system to obtain the standard coordinates of the reference feature points; the coordinates of the matching feature points of the new keyframe are compared with the standard coordinates of the reference feature points; the spatial transformation relationship between the two types of feature point coordinates is solved using the homography matrix; and the transformation relationship is decomposed to obtain the scaling, translation, and rotation parameters used to correct the new keyframe. The scaling parameters are used to eliminate image scale differences caused by airway expansion and contraction; the translation parameters are used to eliminate image position deviations caused by overall airway displacement; and the rotation parameters are used to eliminate image angle deviations caused by slight airway torsion.

[0099] The newly added keyframes are benchmarked according to the correction parameters, so that the corrected new keyframes are aligned with the benchmark keyframes in the feature coordinate system, ensuring that subsequent anomaly analysis is only for changes in airway pathological state, rather than image differences caused by physiological motion.

[0100] Preferably, the step of identifying potentially anomalous frames includes:

[0101] The feature difference index is calculated between the corrected new keyframe and the baseline keyframe to identify potential abnormal frames; the feature difference index includes: color feature difference index, texture feature difference index and contour feature difference index.

[0102] The HSV color space parameters of the corrected new keyframe and the baseline keyframe are extracted separately, including the mean hue, mean saturation, mean lightness, and the variances of hue, saturation, and lightness, to comprehensively reflect the overall characteristics and distribution uniformity of the mucosal color. The color deviation between the new keyframe and the baseline keyframe is calculated using a weighted Euclidean distance algorithm as a color feature difference indicator. The weighting coefficient is set based on the importance of clinical color perception to comprehensively quantify the overall degree of change in mucosal color and accurately capture the hue and saturation shifts caused by abnormal mucosal congestion.

[0103] The texture parameters of the cavity wall and mucosal surface in the corrected new keyframe and the reference keyframe are extracted by the gray-level co-occurrence matrix algorithm, including energy, entropy value and contrast. The absolute difference of the corresponding texture parameters between the new keyframe and the reference keyframe is calculated first. The texture deviation value is obtained by weighted summation as a texture feature difference index, which comprehensively reflects the changes in the roughness of the mucosal surface and the condition of the attached substances.

[0104] The Canny edge detection algorithm is used to extract the edge contours of the corrected new keyframes and the baseline keyframes. The contour parameters such as perimeter, area, equivalent diameter and edge smoothness of the edge contours are further extracted. After normalizing the absolute deviation of each contour parameter, the contour shape deviation value is calculated by weighted summation as the contour feature difference index. The overall degree of change of lung morphology is comprehensively quantified to effectively capture the changes in contour size and regularity caused by abnormal lung morphology.

[0105] Based on statistical data of normal airway characteristics in clinical practice, normal fluctuation thresholds for each characteristic difference index are preset. If the deviation value of any characteristic difference index of a newly added keyframe is greater than the corresponding normal fluctuation threshold, the newly added keyframe is marked as a potential abnormal frame, and the type of characteristic deviation exceeding the normal fluctuation threshold is recorded.

[0106] Please see Figure 5 Preferably, the specific steps for segmenting and quantizing potential anomalous frames to determine anomalous keyframes include:

[0107] Based on the type of feature deviation, feature images of the corrected new keyframe and the reference keyframe are generated respectively. For example, if the feature deviation type is color deviation, the hue component map and saturation component map of the mucosa region are extracted from the corrected new keyframe and the reference keyframe respectively as color feature images. Pixel-level difference operations are performed on the color feature images of the same type of keyframes to obtain color difference images. The difference results highlight the pixel areas with significant color deviation. If the feature deviation type is texture deviation, the entropy values ​​of the corrected new keyframe and the reference keyframe are compared using the gray-level co-occurrence matrix texture mapping algorithm. The texture parameters are converted into grayscale texture images as texture feature images. Pixel-level difference operations are performed on the texture feature images of the same type for the two types of keyframes to obtain texture difference images. The difference results highlight pixel areas with significant texture fluctuations. If the feature deviation type is contour deviation, the Canny edge detection results of the corrected new keyframe and the reference keyframe are converted into edge binary images as contour feature images using an edge binarization processing algorithm. Pixel-level difference operations are performed on the contour feature images of the two types of keyframes to obtain contour difference images. The difference results highlight pixel areas with significant differences in contour shape.

[0108] The feature images of the newly added keyframe and the reference keyframe are differentially processed to generate a difference image. An effective threshold for each pixel in the difference image is determined by an adaptive threshold segmentation algorithm. The effective threshold is automatically calculated based on the gray-level histogram distribution of the difference image to eliminate minor noise interference.

[0109] Pixels with gray values ​​greater than the effective threshold in the difference image are marked as difference pixels. Then, the discrete difference pixels are aggregated into continuous abnormal regions by the region growing algorithm. At the same time, false regions with areas smaller than the preset minimum region area are removed to obtain the abnormal regions of potential abnormal frames.

[0110] The relative area ratio of the abnormal region in a potential abnormal frame is calculated as the degree of abnormality. If the degree of abnormality is greater than the preset area threshold, the potential abnormal frame is determined to be an abnormal key frame. The preset area threshold refers to the minimum relative area standard of an abnormal region that can be clinically identified and may have a substantial impact on surgical safety or the patient's airway ventilation function. It is set by statistical analysis based on a large number of clinical airway abnormality cases, combined with the minimum effective identification size of airway observation during surgery, the degree of impact of the abnormal region on airway ventilation efficiency and surgical operating space, and the matching of device image resolution and clinical visual recognition accuracy.

[0111] Dynamic change analysis of abnormal keyframes quantifies abnormal change trends: Abnormal keyframes of the same characteristic deviation type are concatenated in sequence according to the acquisition timestamp, and the growth rate of the relative area ratio of abnormal regions in adjacent abnormal keyframes is calculated to generate an abnormal change trend curve. Based on the growth rate of the abnormal change trend curve, the abnormal state is determined, including progression, stability, and remission. This dynamic change analysis provides doctors with real-time dynamic information on the development of abnormalities, assisting them in accurately judging whether immediate intervention is needed, maintaining the current observation frequency, or evaluating the effectiveness of intervention operations. This avoids misjudgment or delayed treatment caused by static abnormality judgment, and improves the scientific and timely nature of clinical decision-making.

[0112] Step S5: The intelligent control module establishes a linkage control with the display terminal, constructing a dual-video stream adaptive display mechanism. Based on the degree of abnormality of the abnormal keyframes and a collaborative dataset, the video source of the display terminal is switched. By default, the display area of ​​the control terminal focuses on displaying the cuff video stream and key parameters such as the stent positioning depth and cuff pressure, allowing doctors to monitor the cuff fit and insertion position stability in real time. The cavity video stream is transmitted to the intelligent control module for analysis in the background without occupying display area resources, thus avoiding distraction of the doctor's attention. Adaptive switching is triggered based on the degree of abnormality of the abnormal keyframes. When an abnormal keyframe is detected, a switching command is sent to the display terminal, switching the display area content to the abnormal segment of the cavity video stream, simultaneously overlaying abnormal area markers and abnormal status prompts, and triggering audio-visual prompts to remind the doctor to confirm in time. At the same time, doctors can manually switch between the cuff video stream and the cavity video stream via the intelligent control module's button or an external foot switch, achieving comprehensive observation of the cuff status and cavity abnormalities, balancing clinical operation focus and timely response to abnormalities.

[0113] In this embodiment, step S5 aims to address the conflicting issues of simultaneously displaying dual video streams, which can easily distract doctors and make it difficult to detect cavity abnormalities in a timely manner, while doctors need to simultaneously monitor the cuff status and observe cavity abnormalities. In practice, the default focus is on the cuff video stream and key parameters, ensuring doctors can concentrate on the core operation of cannulation position and cuff fit; in case of abnormalities, the system automatically switches the cavity stream and triggers a prompt, avoiding missed diagnoses and delays; the manual switching function meets the need for comprehensive observation, ultimately achieving a balance between clinical focus and timely response to abnormalities, improving surgical safety and operational efficiency.

[0114] Preferably, the specific steps for switching the video source on the display terminal include:

[0115] The abnormal keyframes identified are classified into different levels of abnormality based on the relative area ratio of the abnormal region and the degree of aggravation of feature deviation. These levels include mild, moderate and severe.

[0116] Based on the severity level of the anomaly, video source switching rules are set to generate switching instructions and send them to the display terminal. In case of severe anomaly, immediate switching is triggered; in case of moderate anomaly, delayed switching is triggered to reserve buffer time for the doctor's current operation; in case of mild anomaly, no display area switching is triggered, only a background prompt is activated; at the same time, a switching instruction data packet containing the coordinates of the abnormal area and the severity of the anomaly is generated and sent to the display terminal.

[0117] After receiving the switching command, the display terminal first pauses the playback of the default video stream in the display area, while retaining the superimposed display of key parameters such as the stent positioning depth and cuff pressure. It then switches the video source of the display area from the cuff video stream channel to the cavity video stream channel, automatically locates the abnormal key frame in the collaborative data set, and supports superimposing magnified images of abnormal areas of the abnormal key frames in the cavity video stream to facilitate doctors' observation of details.

[0118] When a doctor triggers manual switching via the video switching button on the intelligent control module panel or an external foot switch, the display terminal receives the manual switching signal and prioritizes interrupting the current adaptive switching process.

[0119] If the current display area is the endoscopic video stream, it switches to the cavity video stream and overlays an enlarged view of the abnormal area of ​​the abnormal keyframe; if the current display area is the cavity video stream, it switches back to the endoscopic video stream, and the key parameters are overlaid and displayed without interruption in both switching states; after manual switching, the display terminal reports the switching status to the intelligent control module, and the controller records the switching timestamp and operator information and stores them in the collaborative dataset log.

[0120] This embodiment describes a dual-camera visual flushing dual-lumen endotracheal intubation device, including a data acquisition module, an intelligent control module, a keyframe processing module, an anomaly detection module, and a display linkage module. Each module is connected via a data line or a wireless communication link. The specific structure and functions are as follows:

[0121] The data acquisition module includes a dual-camera assembly, a locator, a cuff pressure sensor, an airway environment sensor, a ventilation flow sensor, a device status detector, and supporting structural components. The dual-camera assembly consists of a main camera and a secondary camera, both using LED cameras. The main camera is fixed to the outer wall of the tube near the bronchial cuff and tracheal cuff, while the secondary camera is integrated into the tip and has a cold light source. The tip is covered with a stainless steel sheath and fixed to the inner wall of the bronchial end of the tube. The locator has a built-in displacement detection element embedded in the proximal end of the tube, used to detect the axial displacement of the tube core within the tube. The cuff pressure sensor is attached to the bronchus... The inner walls of the endotracheal cuff and tubal cuff are used to collect the pressure inside the cuff; the airway environment sensor is embedded in the inner wall of the bronchus end of the tubing to collect the temperature and humidity inside the airway; the ventilation flow sensor is connected in series in the ventilation channels of the endotracheal extension tube and bronchial extension tube of the tubing to collect the ventilation flow; the equipment status detector is integrated into the housing of each component, and is also associated with the working status detection of the inflation tube, indicator balloon, one-way valve, flushing tube, and flushing tube connector to detect the component's working voltage, communication status, and the connection status of the flushing and inflation circuits; the supporting structural components include the tubing, tubing connector, and dedicated smart connector, providing a mounting carrier and data transmission interface for each data acquisition element.

[0122] The data acquisition module is used to collect real-time video stream data of the cuff, video stream data of the cavity, core positioning depth data, cuff pressure data, airway environment data, ventilation flow data, and equipment operating status data. When the camera is obscured by sputum or mist, it can be used with the flushing tube and flushing tube connector to flush and aspirate secretions, ensuring the clarity of the collected data. Finally, all the collected raw data is transmitted to the intelligent control module through a dedicated smart connector.

[0123] The intelligent control module integrates a main processor, a data interface unit, an instruction generation unit, and a timing unit. As the core control component of the device, the intelligent control module coordinates the main processor, data interface unit, instruction generation unit, and timing unit, enabling data reception, parsing and processing, logical judgment, and instruction generation. The data interface unit establishes communication with the data acquisition module through a dedicated intelligent connector, and simultaneously establishes data interaction links with the keyframe processing module, anomaly detection module, display linkage module, and storage module. The instruction generation unit generates camera start instructions, cavity video stream sampling analysis trigger instructions, video source switching instructions, and auxiliary instructions for flushing and inflation circuit control. The timing unit provides a unified time reference and calibrates the timestamps of each data acquisition.

[0124] The intelligent control module receives raw data transmitted from the data acquisition module and generates control commands according to preset logic; it performs timestamp calibration and core positioning depth correlation on the cuff video stream data and cavity video stream data to construct a collaborative dataset; it determines whether to trigger cavity video stream sampling analysis operation, sends cavity video stream sampling analysis command to the keyframe processing module, constructs a dual video stream adaptive display mechanism, sends video source switching command to the display linkage module, and simultaneously links the auxiliary control of the inflation tube and flushing tube; it transmits the collaborative dataset and operation log to the storage module through the data interface unit for storage, and coordinates the collaborative operation of each module.

[0125] The keyframe processing module includes a sampling interval adjustment unit, a validity screening unit, and a redundancy removal unit. The sampling interval adjustment unit has built-in interval calculation logic to dynamically adjust the sampling interval based on the inter-frame differences in the cavity video stream. The validity screening unit has a built-in image sharpness detection algorithm to remove blurred frames caused by sputum obscuring or fog interference. The redundancy removal unit has a built-in image similarity comparison algorithm to remove duplicate frames with no substantial differences in airway status.

[0126] The keyframe processing module receives the cavity video stream data and sampling analysis instructions issued by the intelligent control module. It extracts an initial keyframe candidate set according to the initial sampling interval through the sampling interval adjustment unit; it removes fuzzy frames through the validity screening unit and redundant frames through the redundancy removal unit, and finally constructs a keyframe image set with a concise structure and complete information, and transmits it to the anomaly judgment module.

[0127] The anomaly detection module includes a baseline correction unit, a feature difference calculation unit, an anomaly region segmentation and quantization unit, and a dynamic trend analysis unit. The baseline correction unit incorporates feature detection algorithms and homography matrix solving logic to locate baseline keyframes, extract stable feature points, and generate correction parameters. The feature difference calculation unit incorporates HSV color parameters, gray-level co-occurrence matrix texture parameters, and Canny edge contour parameters calculation logic to calculate color, texture, and contour feature difference indices. The anomaly region segmentation and quantization unit incorporates feature image difference logic, an adaptive threshold segmentation algorithm, and a region growing algorithm to segment anomaly regions and calculate their relative area proportions. The dynamic trend analysis unit incorporates growth rate calculation logic to concatenate anomaly keyframes and analyze anomaly change trends.

[0128] The anomaly detection module receives the set of keyframe images transmitted by the keyframe processing module, performs benchmark correction on newly added keyframes through the benchmark correction unit, identifies potential abnormal frames through the feature difference calculation unit, determines and classifies abnormal keyframes through the abnormal region segmentation and quantification unit, and simultaneously judges whether the bronchial cuff positioning is abnormal; quantifies the abnormal change trend through the dynamic trend analysis unit, and transmits the abnormal keyframe information, the abnormality level classification result, and the cuff positioning abnormality prompt to the intelligent control module.

[0129] The display linkage module includes an instruction receiving unit, a video source switching unit, an information overlay unit, an audio-visual prompt unit, and a manual switching unit. The instruction receiving unit receives video source switching instructions from the intelligent control module. The video source switching unit has built-in video channel switching logic for switching between the cuff video stream channel and the cavity video stream channel. The information overlay unit overlays key parameters such as stent positioning depth and cuff pressure, as well as abnormal area markers, abnormal prompt text, and cuff positioning abnormality warnings onto the video stream. The audio-visual prompt unit includes LED indicators and a buzzer to trigger corresponding audio-visual prompts according to the severity of the abnormality. The manual switching unit includes physical buttons and an external foot switch interface for doctors to manually trigger video source switching.

[0130] The display linkage module is used to link with the intelligent control module. By default, the main display area of ​​the control display terminal displays the cuff video stream and key parameters, allowing doctors to monitor the bronchial cuff and tracheal cuff fit status and intubation position stability in real time. After receiving abnormal key frame information, it switches to the abnormal segment of the cavity video stream according to the degree of abnormality, superimposes the abnormal information and triggers audio-visual prompts. Doctors can freely switch between the two video streams through the manual switching unit, and can simultaneously observe the cuff positioning and the internal state of the cavity. The switching status is fed back to the intelligent control module and recorded synchronously to the storage module.

[0131] Working principle and its effects:

[0132] The core working principle of this invention is to use the core positioning depth as the core triggering basis to achieve accurate acquisition, intelligent analysis and adaptive display of dual video streams. Ultimately, it takes into account both clinical operation focus and timely response to airway abnormalities, and solves the pain points of existing technologies such as single data acquisition, delayed abnormality identification and display interference with operation.

[0133] Specifically, after the device is started, it captures the positioning depth of the sclere in real time, accurately triggers the dual cameras to collect video streams from the cuff and cavity, and binds timestamps and positioning information to build a collaborative dataset, ensuring that the data is comprehensive and closely correlated in terms of time and location, avoiding invalid data collection. After the sclere is determined to be in a stable state through the stability monitoring window, the cavity video stream is sampled and analyzed. The sampling interval is dynamically adjusted in combination with the inter-frame difference, and a high-quality keyframe set is constructed through validity screening and redundancy removal, which improves the analysis efficiency and ensures the validity of the data. Physiological motion interference is eliminated through benchmark correction, and potential abnormal frames are identified based on multi-dimensional feature differences. Abnormal keyframes are determined by segmentation and quantification, and the changing trend is analyzed to achieve accurate identification of abnormalities. Finally, the video source of the display terminal is adaptively switched according to the degree of abnormality. The default focus is on the cuff video stream to ensure the operator's focus. When an abnormality occurs, it automatically switches to the cavity abnormal segment and strengthens the prompt. Manual switching is also supported to balance information integrity and operation continuity.

[0134] In summary, this invention not only achieves comprehensive acquisition of airway and cuff status information and accurate and timely identification of airway abnormalities, but also avoids ineffective computational waste and operational interference, significantly improving the safety, operational efficiency, and scientific nature of clinical decision-making in double-lumen endotracheal intubation surgery, and optimizing the overall application effect.

[0135] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A dual camera visualizing dual lumen bronchial intubation device, characterized in that, The data acquisition module is used for real-time acquisition of cuff video stream data, cavity video stream data and tube core positioning depth data; the intelligent control module is used for construction of a collaborative data set, judgment of whether to trigger cavity video stream sampling analysis operation, and construction of a double video stream adaptive display mechanism; the key frame processing module is used for key frame image extraction of cavity video stream data through regulation of sampling interval, and construction of a key frame image set; the abnormality judgment module is used for reference correction of a new key frame in the key frame image set; feature difference index calculation is performed on the corrected new key frame and the reference key frame of the key frame image set to identify potential abnormal frames; and abnormal region segmentation and quantization are performed on the potential abnormal frames to determine abnormal key frames; and the display linkage module is used for adaptive switching of video sources of a display terminal. The intelligent control module is configured with a collaborative control method, and the collaborative control method comprises: Real-time capture of tube core positioning depth of a tube core in a tube body is used to judge whether to trigger real-time video data acquisition, if yes, cuff video stream data and cavity video stream data are collected respectively to construct a collaborative data set; Calculation of change amplitude of tube core positioning depth in a preset monitoring window is used to identify a stable monitoring window to judge whether to trigger cavity video stream sampling analysis operation; When the cavity video stream sampling analysis operation is triggered, key frame image extraction is performed on cavity video stream data of the collaborative data set through regulation of sampling interval, and a key frame image set is constructed in combination with effectiveness screening and redundancy elimination optimization; Reference key frames in the key frame image set are located, stable feature points of the reference key frames are extracted, a feature coordinate system is established to perform reference correction on new key frames in the key frame image set; feature difference index calculation is performed on the corrected new key frames and the reference key frames to identify potential abnormal frames; and abnormal region segmentation and quantization are performed on the potential abnormal frames to determine abnormal key frames; A double video stream adaptive display mechanism is constructed, and video sources of a display terminal are adaptively switched based on the collaborative data set according to the abnormal degree of the abnormal key frames. The step of constructing a collaborative data set comprises: When the double-camera visual flushing double-lumen bronchial intubation device starts to run, the intelligent control module triggers a data acquisition process and activates a data acquisition component; Real-time capture of tube core positioning depth of a tube core in a tube body is used to judge whether to trigger real-time video data acquisition, if yes, cuff video stream data and cavity video stream data are collected respectively to construct a collaborative data set; 2. The dual camera visualizing dual lumen bronchial tube device of claim 1, wherein, A start instruction is issued to double cameras through the intelligent control module, the double cameras comprising a main camera and a secondary camera; the main camera is used to collect cuff video stream data; and the secondary camera is used to collect cavity video stream data; Collection time stamps of the cuff video stream data and the cavity video stream data are extracted respectively, and the time stamps of the cuff video stream data and the cavity video stream data are calibrated and aligned with the internal clock of the intelligent control module as a unified time reference; ​ ​ ​ The tube positioning depth data is bound to the capsule video stream data and the cavity video stream data according to the collection time stamp based on the tube positioning depth, and a cooperative data set is formed.

3. The dual camera visualizing dual lumen bronchial tube device of claim 1, wherein, The step of judging whether to trigger the cavity video stream sampling analysis operation comprises: Real-time receiving the tube positioning depth, sampling the tube positioning depth according to a preset time interval, and constructing a tube positioning depth sequence; Meanwhile, a monitoring window is configured, the sampled tube positioning depth sequence in the monitoring window is filtered and denoised, and the extreme value of the filtered and denoised tube positioning depth sequence is extracted, and the extreme value difference is calculated as the change amplitude of the tube positioning depth in the monitoring window; If the change amplitude is less than a preset fluctuation threshold, the monitoring window is marked as a stable monitoring window; the cumulative duration of the stable monitoring window and the interval duration between adjacent two stable monitoring windows are counted; If the cumulative duration of the stable monitoring window is greater than a minimum stable cumulative duration, and the interval duration of all adjacent stable monitoring windows is less than or equal to a maximum allowed interval duration, it is determined that the tube positioning depth is continuously in a stable state, and the cavity video stream sampling analysis operation is triggered.

4. The dual camera visualizing double lumen bronchial tube device of claim 1, wherein, The specific steps of regulating the sampling interval comprise: An initial sampling interval is set, and after the cavity video stream sampling analysis operation is triggered, key frames are extracted from the cavity video stream data in the cooperative data set according to the initial sampling interval, and an initial key frame candidate set is formed; Difference analysis is performed on the continuous key frame images in the initial key frame candidate set, the gray difference, structural similarity index and feature point matching degree of the corresponding pixel regions of the continuous key frame images are calculated respectively, and the frame difference value is generated through normalization processing; The frame difference value is compared with a preset difference threshold, and the preset difference threshold comprises a preset difference lower threshold and a preset difference upper threshold; if the difference value is less than the preset difference lower threshold, a sampling reduction amplitude is generated according to the deviation degree of the difference value and the preset difference lower threshold, which is used to increase the sampling interval; If the difference value is greater than the preset difference upper threshold, a sampling increase amplitude is generated according to the deviation degree of the difference value and the preset difference upper threshold, which is used to shorten the sampling interval.

5. The dual camera visualizing double lumen bronchial tube device of claim 4, wherein, The step of constructing the key frame image set comprises: The initial key frame candidate set is subjected to effectiveness screening, the gray gradient mean, edge contour sharpness and signal noise ratio of the key frames of the initial key frame candidate set are calculated respectively, and are compared with the corresponding preset standard threshold respectively, and blurred frames are identified and removed; The initial key frame candidate set after effectiveness screening is subjected to redundancy removal optimization, and the image similarity of the continuous key frames is calculated; if the image similarity of the continuous key frames is higher than a preset redundancy threshold, the previous frame key frame is retained as an effective key frame, and the next frame redundant key frame is removed; otherwise, both are retained as effective key frames, and a key frame image set is formed.

6. The dual camera visualizing double lumen bronchial tube device of claim 1, wherein, The step of performing reference correction on the new key frame in the key frame image set comprises: The reference key frame of the key frame image set is located, and the reference key frame is subjected to image preprocessing; The stable feature points of the reference key frame are extracted from the reference key frame after image preprocessing; Generate a binary feature vector of each stable feature point based on the local pixel gray distribution and gradient direction of the stable feature point; Take any stable feature point as a core reference point, calculate the linear distance and relative direction of the remaining stable feature points relative to the core reference point, and determine the relative position relationship of each stable feature point; and take the core reference point as the origin to establish a feature coordinate system; Preprocess the image of the newly added key frame sampled after the reference key frame in the key frame image set, extract the newly added feature points of the newly added key frame, and generate the binary feature vectors of the newly added feature points; Based on the feature vectors of the newly added feature points and the binary feature vectors of the stable feature points of the reference key frame, calculate the feature point similarity, and if the feature point similarity between the newly added feature point and the stable feature point is greater than the similarity threshold, determine that the corresponding newly added feature point and the stable feature point are a matching feature point pair; According to the feature coordinate system, perform spatial coordinate mapping and deviation calculation on the matching feature point pair, solve the homography matrix to generate the correction parameters of the newly added key frame, including the scaling parameter, the translation parameter, and the rotation parameter; and perform reference correction on the newly added key frame according to the correction parameters.

7. The dual camera visualizing double lumen bronchial tube device of claim 6, wherein, The step of identifying the potential abnormal frame includes: Calculate the feature difference index of the corrected newly added key frame and the reference key frame, including the color feature difference index, the texture feature difference index, and the contour feature difference index; The color feature difference index is obtained by calculating the color deviation value of the HSV color space parameters of the corrected newly added key frame and the reference key frame; The texture feature difference index is obtained by calculating the absolute difference value of the texture parameters of the inner wall and mucosa surface of the corrected newly added key frame and the reference key frame; The contour feature difference index is obtained by extracting the contour parameters of the edge contour of the corrected newly added key frame and the reference key frame, and normalizing the absolute deviation of each contour parameter; Respectively preset the normal fluctuation threshold of each feature difference index, if the deviation value of any type of feature difference index of the newly added key frame is greater than the corresponding normal fluctuation threshold, the newly added key frame is marked as a potential abnormal frame, and the feature deviation type that exceeds the normal fluctuation threshold is recorded.

8. The dual camera visualizing double lumen bronchial tube device of claim 7, wherein, The step of segmenting and quantifying the abnormal area of the potential abnormal frame to determine the abnormal key frame includes: According to the feature deviation type, generate the feature images of the corrected newly added key frame and the reference key frame respectively; perform difference operation on the feature images of the newly added key frame and the reference key frame to generate a difference image, and determine the effective threshold of each pixel point in the difference image through an adaptive threshold segmentation algorithm; Mark the pixels with gray value greater than the effective threshold in the difference image as difference pixels, and then aggregate the discrete difference pixels into continuous abnormal areas through a region growing algorithm, while eliminating false areas with an area smaller than a preset minimum area, to obtain the abnormal area of the potential abnormal frame; Calculate the relative area proportion of the abnormal area of the potential abnormal frame, and if it is greater than a preset area threshold, determine that the potential abnormal frame is an abnormal key frame, and perform dynamic change analysis on the abnormal key frame to quantify the abnormal change trend and determine the abnormal state.

9. The dual camera visualizing double lumen bronchial tube device of claim 1, wherein, The step of adaptively switching the video source of the display terminal comprises: grading the identified abnormal key frame according to the degree of abnormality, setting a video source switching rule according to the degree of abnormality, generating a switching instruction, and sending the switching instruction to the display terminal; After receiving the switching instruction, the display terminal pauses the default video stream playing in the display area, switches the video source of the display area from the cuff video stream channel to the cavity video stream channel, and automatically locates to the abnormal key frame in the collaborative data set to superimpose the abnormal area magnified view of the abnormal key frame; When the manual switching is triggered, the display terminal receives the manual switching signal and interrupts the current adaptive switching process: if the current display area is the cuff video stream, the display terminal switches to the cavity video stream and superimposes the abnormal area magnified view of the abnormal key frame; if the current display area is the cavity video stream, the display terminal switches back to the cuff video stream.

Images