ECMO tube blood flow state monitoring method and system based on video analysis

Abstract

The application discloses an ECMO pipeline blood flow state monitoring method and system based on video analysis and belongs to the technical field of medical monitoring. The method comprises the following steps: acquiring a video stream of an ECMO pipeline, performing pipeline center dynamic positioning, and extracting a one-dimensional average gray time sequence signal; calculating a dynamic baseline of the time sequence signal and extracting a normalized waveform event segment; calculating an optical polarity index and a normalized energy dispersion index of the waveform event segment, wherein the optical polarity index represents a positive and negative energy distribution difference, and the normalized energy dispersion index represents energy concentration or dispersion form; and finally, constructing a feature vector based on the optical polarity index and the normalized energy dispersion index, using a classifier to identify bubbles, and realizing monitoring. The application discloses a method for monitoring bubbles in a complex clinical environment, which significantly improves the accuracy and reliability of bubble monitoring.

Description

Technical Field

[0001] This invention relates to the field of medical monitoring technology, specifically to a method and system for monitoring the status of extracorporeal circulation tubing based on video image processing, which is particularly suitable for detecting tiny air bubbles in ECMO (extracorporeal membrane oxygenation) tubing. Background Technology

[0002] Extracorporeal membrane oxygenation (ECMO) is a key technology in modern critical care medicine used to provide continuous extracorporeal cardiopulmonary support. During ECMO operation, the patient's blood is drawn out of the body, flows through a centrifugal pump and oxygenator, and is then returned to the body. The tubing system is the only channel for extracorporeal blood circulation, and its safety is paramount. A major risk during ECMO operation is the introduction of microbubbles. If these bubbles enter the patient's arterial system, they can cause air embolism, resulting in severe neurological damage or even death. Therefore, real-time and accurate monitoring of air bubbles in the ECMO tubing is one of the core requirements for ensuring patient safety.

[0003] In existing technologies, one non-invasive monitoring approach involves using a camera to capture real-time video streams of the ECMO tubing. Image processing techniques are used to create one or more virtual scan lines on the video frames, extracting the pixel intensity of blood and contents (such as air bubbles or blood clots) flowing along these lines. This transforms the two-dimensional image problem into a one-dimensional time-series signal analysis problem. This method attempts to identify air bubbles by analyzing the waveform characteristics of this one-dimensional signal. For example, due to their transparency or reflectivity, air bubbles may produce a bright signal at their edges, while being translucent or absorbing light at their center, potentially resulting in a W-shaped double-peak waveform. In contrast, tiny blood clots or lipid droplets are opaque and primarily absorb light, potentially producing a U-shaped single-valley waveform.

[0004] However, existing technologies based on one-dimensional intensity profile analysis have serious drawbacks in clinical applications: their waveform characteristics are extremely sensitive to changes in ambient light. The lighting conditions in the intensive care unit where ECMO equipment is located are complex and dynamically changing, including variations in natural light during the day, local lighting used during nighttime care, and brief exposure to flashlights during staff rounds. When the intensity or angle of ambient light changes, the baseline brightness and contrast of the tubing in the video drift significantly. This leads to two main problems: First, under low light conditions, the W-shaped waveform produced by bubbles has extremely low amplitude, with peak values ​​potentially even lower than the normal background noise of blood under strong light, causing the system to fail to recognize it and resulting in missed detections; second, under strong reflective light at specific angles, a normal cluster of blood cells or tiny water vapor particles on the tubing may also produce a momentary high-brightness signal, which the system may misinterpret as a bubble, leading to false alarms. Most existing technologies rely on fixed thresholds or waveform templates based on absolute brightness for matching, which cannot adapt to such real-time dynamic changes in lighting, resulting in severely insufficient robustness and reliability of the monitoring system and posing significant clinical safety risks. Therefore, there is an urgent need for a method for monitoring air bubbles in ECMO pipelines that can overcome the interference of ambient light. Summary of the Invention

[0005] This invention provides a video analysis-based method and system for monitoring blood flow status in ECMO tubing, to solve the technical problems that the monitoring results are easily affected by changes in ambient light, have poor robustness, and are prone to missed detections and false alarms due to reliance on absolute brightness analysis.

[0006] In a first aspect, the present invention provides a method for monitoring the blood flow status of ECMO tubing based on video analysis, comprising the following steps: acquiring a video stream of the ECMO tubing, dynamically locating the tubing center in the image frames of the video stream, extracting the average gray value of the scan line at the center of the tubing, and obtaining a one-dimensional average gray time series signal; performing dynamic baseline calculation on the one-dimensional average gray time series signal to obtain a dynamic baseline signal, and extracting a normalized event signal based on the difference between the one-dimensional average gray time series signal and the dynamic baseline signal, segmenting the normalized event signal to obtain waveform event segments; calculating the optical polarity index and the normalized energy dispersion index of the waveform event segments, wherein the optical polarity index is used to characterize the difference in positive and negative energy distribution of the waveform event, and the normalized energy dispersion index is used to characterize the energy concentration or dispersion pattern of the waveform event; constructing a feature vector based on the optical polarity index and the normalized energy dispersion index, and using a preset classifier to discriminate the feature vector to identify whether the waveform event segment corresponds to a bubble, thereby realizing the monitoring of the blood flow status of the ECMO tubing.

[0007] The monitoring method provided by this invention extracts normalized event signals by calculating dynamic baselines and constructs two morphological characteristic indicators, namely the optical polarity index and the normalized energy dispersion index, which are robust to changes in illumination and flow velocity. It eliminates the dependence on absolute signal intensity and can effectively overcome the influence of ambient light interference and changes in blood flow velocity, significantly improving the accuracy and reliability of monitoring microbubbles in complex clinical environments.

[0008] Furthermore, the dynamic positioning of the pipeline center in the image frames of the video stream includes: performing edge detection on the image frames and extracting the edges on both sides of the pipeline; performing projection analysis on the edges on both sides of the pipeline to determine the positions of two valleys in the projection histogram; and calculating the midpoint of the two valley positions as the centerline position of the pipeline in the current image frame.

[0009] By dynamically locating the center of the tubing in each frame of the image, the scan line is always aligned with the core area of ​​the tubing, effectively avoiding problems caused by physical vibration of the tubing or patient displacement, and ensuring the continuity and stability of signal acquisition.

[0010] Further, dynamic baseline calculation is performed on the one-dimensional average gray-scale time series signal, including: performing moving median filtering on the one-dimensional average gray-scale time series signal using a sliding window of preset width to obtain the dynamic baseline signal.

[0011] The moving median filter is used instead of the mean filter because the median filter is robust to short-lived, dramatic event waveforms (such as bubble spikes) and will not be disturbed by these "outliers." It can more accurately track the true brightness baseline of a pure blood background, providing an accurate reference for subsequent normalization processing.

[0012] Further, the normalized event signal is segmented to obtain waveform event segments, including: calculating the noise standard deviation of the normalized event signal and setting a dynamic detection threshold based on the noise standard deviation; traversing the normalized event signal, marking the event start time when the absolute value of the signal exceeds the dynamic detection threshold, and marking the event end time when the absolute value of the signal returns to below the dynamic detection threshold, thereby segmenting the waveform event segments.

[0013] Using a dynamic threshold based on the noise standard deviation for event segmentation, compared to a fixed threshold, can better adapt to different devices or background noise levels, ensuring the sensitivity and accuracy of event segmentation and providing complete waveform data for subsequent feature extraction.

[0014] Furthermore, the formula for calculating the optical polarity index is as follows:

[0015] ;

[0016] in, The optical polarity index is... It is the sum of the signal strengths of all signals above the dynamic baseline in the waveform event segment. It is the sum of the absolute values ​​of all signal strengths below the dynamic baseline signal in the waveform event segment. To prevent robustness constants with a denominator of zero.

[0017] The optical polarity index is a normalized index ranging from -1 to +1. The bubble (W-shaped) is caused by its bright edges. Larger, index approaching +1; clots (U-shaped) are caused by shading. The value is relatively large, and the exponent approaches -1. When the light intensity changes, and They scale proportionally, but their ratio (i.e., OPI) remains highly stable, achieving insensitivity to changes in illumination.

[0018] Furthermore, the formula for calculating the normalized energy dispersion index is as follows:

[0019] ;

[0020] in, The normalized energy dispersion index is... The total duration of the waveform event segment. Let be the normalized event signal at time t. The moment the event begins. The moment the event ends. The energy centroid of the waveform event segment is... The total energy of the waveform event segment.

[0021] The Normalized Energy Dispersion Index (NED) draws inspiration from the concept of "moment of inertia." For clumps (U-shaped), energy is concentrated at the center, with an NED value approaching 0; for bubbles (W-shaped), energy is dispersed on both sides, resulting in an NED value significantly greater than 0. This is achieved by dividing by the total duration. Normalize by the square of the value, so that when the flow velocity changes, the waveform width is affected. When the flow rate changes, the numerator and denominator scale proportionally, and the final NED value remains unchanged, thus achieving robustness to changes in flow rate.

[0022] Furthermore, the step of using a preset classifier to discriminate the feature vector includes: using a support vector machine classifier to classify and discriminate the feature vector.

[0023] Furthermore, the method also includes: when the waveform event segment is identified as corresponding to a bubble, triggering an audible and visual alarm, and storing the data and timestamp of the waveform event segment.

[0024] Secondly, the present invention also provides a video analysis-based ECMO tubing blood flow status monitoring system, comprising: a video acquisition module for acquiring a video stream of the ECMO tubing; a signal extraction module for dynamically locating the tubing center in image frames of the video stream, extracting the average grayscale value of the center scan line of the tubing, and obtaining a one-dimensional average grayscale time series signal; and an event analysis module for performing dynamic baseline calculation on the one-dimensional average grayscale time series signal to obtain a dynamic baseline signal, and extracting a normalized event signal based on the difference between the one-dimensional average grayscale time series signal and the dynamic baseline signal, and performing normalized event analysis on the normalized event signal. The event signal is segmented to obtain independent waveform event segments; a feature construction module is used to calculate the optical polarity index and normalized energy dispersion index of the waveform event segments. The optical polarity index is used to characterize the difference in positive and negative energy distribution of the waveform event, and the normalized energy dispersion index is used to characterize the energy concentration or dispersion pattern of the waveform event; a classification decision module is used to construct feature vectors based on the optical polarity index and the normalized energy dispersion index, and use a preset classifier to judge the feature vectors to identify whether the waveform event segment corresponds to a bubble, thereby realizing the monitoring of the blood flow status of the ECMO tubing.

[0025] Thirdly, the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a video analysis-based method for monitoring the blood flow status of ECMO tubing.

[0026] The beneficial effects of this invention are as follows: By reducing the computationally intensive two-dimensional video analysis to one-dimensional time-series signal processing and combining it with moving median filtering to calculate the dynamic baseline, this invention effectively smooths pixel noise and resists low-frequency illumination background drift. The dynamic pipeline positioning function in this invention ensures accurate signal acquisition even when the pipeline experiences physical vibrations, enhancing the system's stability in practical applications. The core of this invention lies in the construction of the Optical Polarity Index (OPI) and the Normalized Energy Dispersion Index (NED). OPI utilizes the essential difference between bubbles and clots in their positive and negative energy components to achieve illumination invariance; NED utilizes the difference in energy distribution patterns and achieves flow velocity invariance through flow velocity normalization. The combination of these two orthogonal robust features enables the system to efficiently and accurately distinguish between microbubbles and blood clots in the complex environment of the ICU, greatly reducing the false negative and false alarm rates and improving the safety of ECMO operation. Attached Figure Description

[0027] Figure 1 This is a schematic diagram of the original signal and typical waveform analysis of the dynamic baseline of the bubble event in an embodiment of the present invention;

[0028] Figure 2This is a schematic diagram of the W-shaped waveform after normalization of the bubble event in an embodiment of the present invention;

[0029] Figure 3 This is a schematic diagram of the original signal and typical waveform analysis of the dynamic baseline of the clot event in an embodiment of the present invention;

[0030] Figure 4 This is a schematic diagram of the normalized U-shaped waveform of the clot event in an embodiment of the present invention;

[0031] Figure 5 This is a schematic diagram illustrating the effect of existing technical features on the feature space under interference;

[0032] Figure 6 This is a schematic diagram illustrating the effect of the feature space under interference in an embodiment of the present invention. Detailed Implementation

[0033] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0034] This invention provides a video analysis-based method for monitoring blood flow status in ECMO tubing, the specific process of which includes the following steps:

[0035] S101: Acquire the video stream of the ECMO tubing, dynamically locate the center of the tubing in the image frames of the video stream, extract the average gray value of the scan line at the center of the tubing, and obtain a one-dimensional average gray time series signal.

[0036] An industrial camera is fixedly installed at a critical location in the ECMO tubing (e.g., at the centrifugal pump outlet). To reduce data processing load and eliminate chromaticity interference, a high frame rate (e.g., 100 frames / second) grayscale camera is preferred, and a stable active light source (such as an infrared LED) can be equipped to initially reduce ambient light interference. The camera acquires a grayscale video stream of the tubing in real time. ,in and These are pixel coordinates. It is frame order.

[0037] To address minor physical displacements in the tubing that may occur due to mechanical vibrations from the centrifugal pump or movement during patient care, this step involves analyzing each frame of the image. Adaptive localization is performed. First, the current frame image is Gaussian blurred to suppress high-frequency noise. Next, the Canny edge detection operator is used to roughly extract the left and right edges of the pipeline. Considering that ECMO pipelines typically have a relatively fixed vertical or horizontal orientation in the video, this embodiment uses a vertical projection method. For example, for a vertical pipeline, the image is pixel-accumulated and projected horizontally to obtain a one-dimensional projection histogram. This histogram will show two distinct troughs due to the pipeline edges. By searching for the positions of these two troughs and calculating their midpoint, the location of the pipeline in the current frame can be dynamically determined. Center line position of the frame In other embodiments, horizontal projection may also be used.

[0038] On the dynamically determined centerline A virtual horizontal scan line is set at this point. This scan line crosses the pipe, and it is assumed that the number of effective pixels within the pipe is [number missing]. Extract all data along the scan line. Collect the grayscale values ​​of each pixel and calculate its average grayscale value. This serves as a one-dimensional profile signal at that moment:

[0039]

[0040] In this way, by using dynamic positioning and mean extraction, the two-dimensional video stream is reduced to a one-dimensional time series signal, which not only reduces the computational complexity, but also smooths out the random noise of individual pixels through mean extraction, while ensuring that the signal acquisition is not affected by the physical jitter of the pipeline.

[0041] S102: Perform dynamic baseline calculation on the one-dimensional average gray-scale time series signal to obtain a dynamic baseline signal, and extract a normalized event signal based on the difference between the one-dimensional average gray-scale time series signal and the dynamic baseline signal. Segment the normalized event signal to obtain independent waveform event segments.

[0042] Extracted signal It includes high-frequency events (i.e., the flow of objects such as bubbles and clots) and low-frequency background information (i.e., pure blood flow). Changes in ambient lighting are mainly reflected in the drift of the low-frequency background. To eliminate this drift, a larger sliding window (e.g., (A frame, the width of which is much larger than the duration of a single event) Perform moving median filtering to obtain the dynamic baseline signal. .like Figure 1 and Figure 3The red dashed line in the image shows the median value instead of the mean. The reason for choosing the median value over the mean value is that the mean filter is distorted by transient, dramatic event waveforms (such as the sudden appearance of bubbles), while the median filter is robust to this and can more accurately track the true brightness baseline of a pure blood background.

[0043] calculate Relative to dynamic baseline Normalized event signal :

[0044]

[0045] at this time, The signal's mean is constant around 0; it only reflects the relative brightness change of the blood flowing through the object relative to the background. For example... Figure 2 and Figure 4 As shown.

[0046] In order to start from continuous Independent event waveforms are segmented from the signal. First, calculations are performed when the system is running stably (i.e., when there are no events). The standard deviation of is denoted as the noise standard deviation. Set a dynamic detection threshold. (Based on the 3-sigma principle). A state machine is used for traversal. :when continuous Frames (e.g.) (Frame) Breaking the threshold upwards When this occurs, it is marked as a "waveform event". The beginning moment When the signal returns to And continue Frames (e.g.) When a frame is created (to prevent the event from being incorrectly truncated due to zero crossings within the waveform), the end time of the event is marked. To capture complete waveform edge information, pre- and post-buffering expansion is performed during segmentation, resulting in actual segmented waveform segments. for exist Signals within a time period, of which The number of frames to buffer (e.g., 10 frames).

[0047] Thus, by using dynamic baseline and normalization processing, low-frequency drift caused by changes in illumination is eliminated, keeping the signal mean constant near 0; by using dynamic threshold segmentation, the complete event waveform is accurately separated from the noise, providing reliable data input for subsequent morphological feature analysis.

[0048] S103: Calculate the optical polarity index and normalized energy dispersion index of the waveform event segment. The optical polarity index is used to characterize the difference between the positive and negative energy distribution of the waveform event, and the normalized energy dispersion index is used to characterize the energy concentration or dispersion pattern of the waveform event.

[0049] For each waveform event segment segmented in S102, calculate the following two illumination invariance indices:

[0050] The Optical Polarity Index (OPI) is used to distinguish the optical physical properties of objects. Microbubbles, due to their spherical shape, produce complex refraction and reflection of light, resulting in waveforms (W-shaped, such as...). Figure 2 Typically, blood clots contain a significant "positive component" (highlight) at the edges and a "negative component" (dark area) at the center. Tiny blood clots or lipid droplets, on the other hand, are opaque, primarily exhibiting light-blocking properties, and their waveform (U-shaped, such as...) Figure 4 It contains almost only "negative components". The OPI index is designed to assess the difference in the distribution of this positive and negative energy.

[0051] First, calculate the total positive energy of the waveform event. Total negative energy :

[0052]

[0053] Based on this, the optical polarity index is defined. as follows:

[0054]

[0055] in, It is a constant set to increase robustness (e.g.) ), used to prevent when and When all values ​​are close to 0 (e.g., a very weak noise fluctuation is incorrectly segmented into an event), the denominator will be 0 or the result will oscillate violently.

[0056] For example, suppose a bubble waveform is segmented, and its normalized signal... It contains two positive peaks and one negative trough. Calculate all peaks above the baseline ( The sum of the signal strengths of the parts is obtained Calculate all values ​​below the baseline ( The sum of the absolute values ​​of the signal strength of the ) parts is obtained Assuming ,but This value is greater than 0. Assuming a clot waveform is segmented, there are almost no positive values. However, there is a deep negative valley. .but The value approaches -1.

[0057] When ambient light changes, the original signal and dynamic baseline They will translate in the same direction or scale proportionally. Because... The difference between the two is that the basic shape of the waveform and the zero point position remain unchanged. and They scale proportionally, but their ratio (i.e., the OPI index) remains highly stable, thus achieving insensitivity to changes in illumination.

[0058] The Normalized Energy Dispersion Index (NED) is used to assess whether the energy of a waveform is concentrated at the center or dispersed on both sides, and the index is designed to be independent of the flow velocity of the object (i.e., the waveform width).

[0059] First, calculate the total duration of the waveform. and total energy Next, the energy centroid of the waveform is calculated. :

[0060]

[0061] Then, the normalized energy dispersion index is defined. as follows:

[0062]

[0063] For example, suppose there is a clot waveform (U-shaped). Frame. Its energy Highly concentrated at the center of the trough, assuming the calculated centroid Because energy is mainly concentrated in nearby, The weighted sum of this term (i.e., the moments of inertia in the molecule) is very small, for example, 4000. Total energy (assumption) ).but .

[0064] For example, suppose there is a bubble waveform (W-shaped). Also 20 frames. Energy Mainly distributed in areas far from the center of mass At the two side peaks (e.g.) and (Location). Therefore, The weighted sum of the terms is very large, for example, 15000. Total energy. (assumption) ).but It can be seen that 0.469 is significantly greater than 0.161, and the NED value can effectively distinguish morphology.

[0065] When the blood flow velocity is reduced by half, the time it takes for the object to flow across the scan line It will become twice the original size (i.e.) At the same time, its center of mass and the time difference at each point It will also be magnified proportionally by 2 times. This will cause the molecules to... The term is magnified by a factor of 4. And the outermost layer of the formula is divided by... This factor is also magnified by a factor of 4. The two cancel each other out, making the NED index independent of the flow velocity (wavelength).

[0066] Thus, by constructing two morphological indices, OPI and NED, which are both illumination-invariant and flow-velocity-invariant, robust feature inputs are provided for subsequent accurate classification.

[0067] S104: Construct a feature vector based on the optical polarity index and the normalized energy dispersion index, and use a preset classifier to judge the feature vector to identify whether the waveform event segment corresponds to a microbubble, thereby realizing the monitoring of the blood flow status of the ECMO tubing.

[0068] For each waveform event calculated in S103, construct a two-dimensional feature vector. .

[0069] Subsequently, the feature vector is discriminated using a pre-defined classifier. This classifier can be trained offline in advance: on an ECMO simulation platform, videos containing known microbubbles, microclots, and lipid droplets are acquired under different lighting and flow rates, thousands of event waveforms are extracted, and their values ​​are calculated. Feature vectors. In the OPI-NED feature space, bubbles will cluster in regions where (OPI>0, NED>0); clots will cluster in regions where (OPI<0, NED≈0). An optimal classification hyperplane is trained using these labeled samples, leveraging the good generalization ability of a support vector machine classifier in low-dimensional space.

[0070] In real-time monitoring, the feature vector calculated by S103 The data is immediately fed into a pre-trained support vector machine classifier for discrimination. Once... If the signal is classified as a "dangerous bubble", the system will immediately trigger an audible and visual alarm and store the waveform data and timestamp as an event record for medical staff to review.

[0071] Figure 5This demonstrates the effects of existing technical features (absolute peak vs. duration) under random illumination and flow rate disturbances. The blue "bubble (W-shaped)" clusters and the green "clump (U-shaped)" clusters are severely stretched and smeared due to random variations in illumination and flow rate, resulting in significant overlap between the two clusters in the feature space, making them difficult to distinguish. In contrast, Figure 6 This demonstrates the effects of the features of the present invention (Optical Polarity Index (OPI) vs. Normalized Energy Dispersion Index (NED)). Despite experiencing [various challenges / advantages], [the invention achieves its intended effect]. Figure 5 Under identical random illumination and flow velocity disturbances, the blue "bubble (W-shaped)" clusters (labeled in the OPI>0 region) and the green "clump (U-shaped)" clusters (labeled in the OPI<0 region) are tightly clustered and clearly separated in the feature space. This intuitively demonstrates that the OPI and NED indices constructed in this invention successfully overcome the disturbances caused by changes in illumination and flow velocity, exhibiting extremely high robustness.

[0072] Thus, by making classification decisions in the robust feature space of OPI-NED, accurate identification of microbubbles was achieved, effectively overcoming the interference of light and flow rate, and ultimately achieving reliable monitoring of the blood flow status of ECMO tubing.

[0073] In another embodiment, the present invention also provides a video analysis-based ECMO tubing blood flow status monitoring system, comprising: a video acquisition module for acquiring a video stream of the ECMO tubing; a signal extraction module for dynamically locating the tubing center in image frames of the video stream, extracting the average grayscale value of the center scan line of the tubing, and obtaining a one-dimensional average grayscale time series signal; and an event analysis module for performing dynamic baseline calculation on the one-dimensional average grayscale time series signal to obtain a dynamic baseline signal, and extracting a normalized event signal based on the difference between the one-dimensional average grayscale time series signal and the dynamic baseline signal, and performing event analysis on the normalized event signal. The event signal is segmented to obtain independent waveform event segments; a feature construction module is used to calculate the optical polarity index and normalized energy dispersion index of the waveform event segments. The optical polarity index is used to characterize the difference in positive and negative energy distribution of the waveform event, and the normalized energy dispersion index is used to characterize the energy concentration or dispersion pattern of the waveform event; a classification decision module is used to construct feature vectors based on the optical polarity index and the normalized energy dispersion index, and use a preset classifier to judge the feature vectors to identify whether the waveform event segment corresponds to a bubble, thereby realizing the monitoring of the blood flow status of the ECMO tubing.

[0074] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the ECMO tubing blood flow status monitoring method based on video analysis in the above embodiments.

Claims

1. A method for monitoring blood flow status in ECMO tubing based on video analysis, characterized in that, include: The video stream of the ECMO tubing is acquired, and the center of the tubing is dynamically located in the image frames of the video stream. The average gray value of the scan line at the center of the tubing is extracted to obtain a one-dimensional average gray time series signal. Dynamic baseline calculation is performed on the one-dimensional average gray-scale time series signal to obtain a dynamic baseline signal. Normalized event signals are extracted based on the difference between the one-dimensional average gray-scale time series signal and the dynamic baseline signal. The normalized event signals are then segmented to obtain waveform event segments. The dynamic baseline calculation for the one-dimensional average gray-scale time series signal includes: performing a moving median filter on the one-dimensional average gray-scale time series signal using a sliding window of a preset width to obtain the dynamic baseline signal; Calculate the optical polarity index and normalized energy dispersion index of the waveform event segment. The optical polarity index is used to characterize the difference between the positive and negative energy distribution of the waveform event, and the normalized energy dispersion index is used to characterize the energy concentration or dispersion pattern of the waveform event. A feature vector is constructed based on the optical polarity index and the normalized energy dispersion index. A preset classifier is used to judge the feature vector to identify whether the waveform event segment corresponds to a bubble, thereby realizing the monitoring of the blood flow status of the ECMO tubing.

2. The method according to claim 1, characterized in that, Dynamically locate the pipeline center of image frames in the video stream, including: Edge detection is performed on the image frame to extract the edges on both sides of the pipeline; Projection analysis is performed on both sides of the pipeline to determine the locations of the two troughs in the projection histogram. Calculate the midpoint between the two trough positions and use it as the centerline position of the pipeline in the current image frame.

3. The method according to claim 1, characterized in that, The normalized event signal is segmented to obtain waveform event segments, including: Calculate the noise standard deviation of the normalized event signal, and set a dynamic detection threshold based on the noise standard deviation; The normalized event signal is traversed, and when the absolute value of the signal exceeds the dynamic detection threshold, it is marked as the start time of the event. When the absolute value of the signal returns to below the dynamic detection threshold, it is marked as the end time of the event, thereby segmenting the waveform event segment.

4. The method according to claim 1, characterized in that, The formula for calculating the optical polarity index is as follows: ; in, The optical polarity index is... It is the sum of the signal strengths of all signals above the dynamic baseline in the waveform event segment. It is the sum of the absolute values ​​of all signal strengths below the dynamic baseline signal in the waveform event segment. To prevent robustness constants with a denominator of zero.

5. The method according to claim 1, characterized in that, The formula for calculating the normalized energy dispersion index is as follows: ; in, The normalized energy dispersion index is... The total duration of the waveform event segment. Let be the normalized event signal at time t. The moment the event begins. The moment the event ends. The energy centroid of the waveform event segment is... The total energy of the waveform event segment; 。 6. The method according to claim 1, characterized in that, The step of using a preset classifier to discriminate the feature vector includes: The feature vectors are classified and judged using a support vector machine classifier.

7. The method according to claim 6, characterized in that, The method further includes: When the waveform event segment is identified as corresponding to a bubble, an audible and visual alarm is triggered, and the data and timestamp of the waveform event segment are stored.

8. A system for implementing the video analysis-based ECMO tubing blood flow status monitoring method according to any one of claims 1 to 7, characterized in that, include: The video acquisition module is used to acquire the video stream from the ECMO tubing. The signal extraction module is used to dynamically locate the center of the pipeline in the image frames of the video stream, extract the average gray value of the scan line at the center of the pipeline, and obtain a one-dimensional average gray time series signal. The event analysis module is used to perform dynamic baseline calculation on the one-dimensional average gray-scale time series signal to obtain a dynamic baseline signal, and extract a normalized event signal based on the difference between the one-dimensional average gray-scale time series signal and the dynamic baseline signal, and segment the normalized event signal to obtain independent waveform event segments. The feature construction module is used to calculate the optical polarity index and normalized energy dispersion index of the waveform event segment. The optical polarity index is used to characterize the difference between the positive and negative energy distribution of the waveform event, and the normalized energy dispersion index is used to characterize the energy concentration or dispersion pattern of the waveform event. The classification decision module is used to construct a feature vector based on the optical polarity index and the normalized energy dispersion index, and to use a preset classifier to judge the feature vector to identify whether the waveform event segment corresponds to a bubble, thereby realizing the monitoring of the blood flow status of the ECMO tubing.

9. A computer-readable storage medium having a computer program stored thereon, the program being executed by a processor to implement the video analysis-based ECMO tubing blood flow status monitoring method as described in any one of claims 1 to 7.

Images