Left beam branch pace-making multi-parameter fusion early warning method based on continuous pace-making mapping method

Abstract

The invention discloses a left beam branch pace-making multi-parameter fusion early warning method based on a continuous pace-making mapping method, and the method comprises the steps: S1, extracting an electrocardiosignal image, and carrying out the ROI region interception and graying processing; s2, analyzing and detecting image tearing, and outputting geometrically consistent electrocardiosignal images through splicing and compensation; s3, each lead signal track is obtained through separation, and a time sequence signal sequence is generated; s4, detecting premature beat / pace-making heart beat and positioning a pace-making stimulation starting point, and outputting a pace-making time sequence; s5, detecting an R wave peak value, and calculating an S-V6RWP value and an S wave depth; s6, in the LBB-I lead signal, determining a base line and positioning a peak value, and calculating an amplitude value; and S7, constructing a fusion early warning model. The system has the advantages that automatic extraction and multi-parameter fusion early warning can be carried out, real-time, automatic and quantitative analysis of central electric signals in the continuous pacing mapping process is achieved, and doctors are assisted in making judgment.

Description

Technical Field

[0001] This invention relates to the field of medical image processing technology, specifically to a multi-parameter fusion early warning method for left bundle branch pacing based on continuous pacing mapping. Background Technology

[0002] Left bundle branch pacing (LBBP) has become a hot topic in cardiac pacing research because it closely approximates the heart's natural conduction pathway, avoiding the drawbacks of right ventricular pacing while providing cardiac resynchronization therapy for heart failure patients. During LBBP surgery, the assessment of the pacing site, especially the precise location and capture of the left bundle branch, is crucial for surgical success and postoperative efficacy. However, traditional intraoperative assessment relies primarily on the physician's manual interpretation and experience of surface and intracardiac electrocardiograms (ECG and EGM), which suffers from high subjectivity, insufficient quantification, and limited assessment efficiency. Therefore, developing an auxiliary diagnostic system capable of real-time, automatic analysis of pacing ECG signals and providing intelligent early warning has significant clinical implications and application value.

[0003] To achieve the above goals, existing technical solutions mainly focus on the automatic analysis of electrocardiogram (ECG) signals, but all face a series of insurmountable bottlenecks. Firstly, there's data acquisition. Ideally, the digital waveform of the ECG signal would be directly acquired through a data interface. However, in reality, the data interfaces of existing hospital-used MEAs, especially older models still in service, are often closed systems with undisclosed communication protocols, making direct acquisition of digital signals impossible. Furthermore, data formats differ between different MEAs. Another approach is to use computer vision technology to directly analyze images or video streams on the MEA screen. However, the age of hospital-used MEAs and their display modules or video output ports can cause serious quality problems such as tearing, misalignment, and jitter in the acquired images due to aging, electromagnetic interference, or graphics card issues. Existing visual algorithms typically assume the input image is complete and clear; such image degradation directly leads to signal trajectory extraction failure or introduces significant errors. Simultaneously, multi-lead ECG signals often overlap during display, especially in cases of complex arrhythmias, making visual separation difficult and resulting in impure extraction of single-lead signals and distorted analysis results. Secondly, after acquiring the ECG signal, existing automated analysis methods still cannot functionally meet the core clinical needs of left bundle branch pacing (LBPB): 1. Insufficient accuracy in detecting the pacing stimulation initiation point: One of the core evaluation indicators of LBPB is the time from the pacing stimulation signal (a high and narrow pulse - S) to the peak R wave in lead V6 of the ECG (S-V6RWPT). Traditional ECG analysis algorithms focus on the QRS complex and lack specific optimization for pacing pulse detection, making them prone to missed or false detections in noisy environments, leading to errors in S-V6RWPT calculation; 2. Lack of a multi-parameter fusion-based early warning system: Current research mainly focuses on the abnormal detection of a single parameter. Intraoperative assessment of LBPB is a multi-dimensional decision-making process that requires the comprehensive evaluation of multiple parameters, such as whether S-V6RWPT is shortened, whether the S wave in lead V6 changes from absent to prominent or from present to deepening, and whether the coronary injury current (COI) (characterized by the amplitude of the pacing potential in lead LBB-I) reaches a safety warning level. Current technology lacks a comprehensive early warning model that organically integrates these parameters and classifies them into different levels and severity.

[0004] In summary, the current technological field lacks a solution that can effectively overcome image quality issues, accurately detect pacing-specific parameters, and perform multi-parameter fusion for intelligent early warning. Summary of the Invention

[0005] The purpose of this invention is to overcome the above-mentioned deficiencies and to disclose to the public a multi-parameter fusion early warning method for left bundle branch pacing based on continuous pacing mapping. Through image compensation, multi-lead separation, automatic extraction of pacing parameters and multi-parameter fusion early warning, it realizes real-time, automatic and quantitative analysis of ECG signals during continuous pacing mapping, and assists doctors in making judgments.

[0006] The technical solution of this invention is implemented as follows: A multi-parameter fusion early warning method for left bundle branch pacing based on continuous pacing mapping includes the following steps: Step S1: Extract ECG signal images frame by frame from the MEA video stream, and perform ROI region cropping and grayscale processing; Step S2: Detect image tearing through continuous inter-frame difference analysis, compensate using piecewise linear stitching and pixel interpolation, and output a geometrically consistent electrocardiogram signal image; Step S3: Apply color mask and morphological closing operation to separate the ECG-V6 lead signal, EGM-LBB lead signal and EGM-LBB-I lead signal trajectory, and generate the time signal sequence of each lead signal by scanning pixels; Step S4: Based on the signals from each lead, detect premature beats / pacing heartbeats and locate the pacing stimulation start point by using first-order difference, Teager energy operator and adaptive double threshold method, and output the pacing time sequence; Step S5: In the ECG-V6 lead signal, based on the pacing time, detect the R wave peak value, calculate the S-V6RWP value and S wave depth; Step S6: In the EGM-LBB-I lead signal, determine the baseline and locate the peak value, and calculate the amplitude; Step S7: Construct a fusion early warning model based on the extracted parameters, including S-V6RWP value, S-wave depth evolution analysis of ECG-V6 lead signal and COI early warning of EGM-LBB-I lead signal, and provide visual early warning through graphical interface and voice prompts.

[0007] In step S1, fixed coordinate interception technology is used to achieve area positioning. The signal display area in the MEA screen is locked by preset ROI parameters, and ROI area interception is performed.

[0008] In step S1, the specific process of grayscale processing is as follows: the weighted average method is used to calculate the pixel grayscale value, which reduces the data dimension while maintaining the contrast between the signal and the background.

[0009] In step S2, the image tearing detection and compensation process includes the following steps: Step (1), consecutive frame difference analysis: The system maintains a double-frame buffer to store the current frame. And the previous frame The grayscale image is used to identify the effective signal region of the two frames: the non-zero state of each row of pixels is calculated, and a binary vector is generated. and Non-zero lines are marked as 1, otherwise as 0; The common valid line region of the two frames is obtained through a logical AND operation: Determine the common valid line range of two frames. If the common valid line range is less than 60% of the image height or the starting line exceeds 40% of the image height, the image quality is considered poor, and the current frame is skipped. Step (2), Displacement Extraction: Within the effective row range, select the first and last two typical signal lines for displacement calculation. For each row, calculate the displacement d between the current frame and the corresponding row of the previous frame; Step (3), Boundary Tear Location: By comparing the displacement of the first and last rows. and Determine if there is tearing in the image: Step a, if If so, it is considered that a rift exists, and the compensation process begins. To set a threshold; Step b: Define a horizontal window, with a width defined as [value to be filled in]. and 2.5 times the maximum value; Step c: Within the horizontal window, calculate the difference between corresponding rows of two frames to determine the location of the tear boundary. The calculation process is as follows: in, The difference threshold, and It refers to the starting position and width of the window area; Step (4), Image Compensation: Based on the position of the torn boundary, divide the image into upper and lower halves, and adjust the displacement of the upper and lower halves accordingly. and The images are translated separately, and a zeroing region is set near the tear point to ensure a smooth transition. The translated upper and lower halves of the image and the zeroing region are then stitched together to obtain the compensated image.

[0010] In step (2), the calculation process for the displacement d is as follows: Step 1) Set This refers to a line of signal from the previous frame. This is the corresponding line signal for the current frame; Step 2) Find the set of pixel locations that satisfy the following conditions: , That is, the position where the current frame signal value is greater than the previous frame and the previous frame signal value was zero; Step 3) If Y is not empty, then start from the first element of Y. Begin by finding the first satisfaction. The location, which is related to The difference is the displacement d. .

[0011] In step S3, the process of separating the signals of each lead is as follows: First, the compensated RGB image obtained in step S2 is converted to the HSV color space. Then, a lead color feature model is established using three independent dimensions: hue, saturation, and brightness. The system has a built-in standard lead color database, which contains the HSV threshold ranges of ECG-V6 lead signals, EGM-LBB lead signals, and EGM-LBB-I lead signals. Initial binary masks for each lead signal are generated through color thresholding. Then, morphological operations are applied to optimize the segmentation results. A dilation-erosion closing operation is used to fill the small gaps in the signal trajectory and connect the broken areas caused by noise. At the same time, an opening operation is used to eliminate isolated noise points. The system analyzes the binary mask column by column from left to right, records the top position coordinates of the signal trajectory in each column, and uses a context-information-based interpolation filling algorithm for columns with missing signals. The reconstructed signal trajectory is then smoothed and filtered to eliminate high-frequency noise introduced during acquisition and segmentation, while preserving the original morphological features of the ECG signal.

[0012] In step S4, the process of locating the pacing stimulation initiation point is as follows: Step a) Signal mutation point detection and feature enhancement: Calculate the first-order difference of the ECG signal sequence, and apply the Teager energy operator (TEO) to enhance the transient characteristics of the signal. The calculation method is to subtract the product of the adjacent difference values ​​from the square of the current difference value, and divide the result by 20 for amplitude limiting, and limit values ​​exceeding 200 to 200; select potential pacing stimulation initiation candidate positions based on the TEO threshold being greater than 20, and at the same time, combine the significant peak points with amplitudes exceeding 700 in the EGM-LBB-I lead signal for spatiotemporal correlation verification, set a matching tolerance range of 50 sampling points, and establish candidate point markers; Step b) Fine screening and confirmation of pacemakers: For each candidate pacemaker, analyze the signal characteristics within the preceding 10 sampling point windows, including calculating the standard deviation of the original EGM-LBB lead signal and ECG-V6 lead signal to verify signal stability. If the standard deviation is greater than a set threshold, it is determined as a non-verification candidate. For non-verification candidate points, the signal baseline is determined using histogram statistics, the difference between the signal and the baseline is calculated, and continuous significant deviation segments are identified. The deviation judgment conditions are set as follows: the number of consecutive deviation sampling points is greater than 2 and the cumulative deviation amplitude exceeds 2, or it is close to the end of the window and the cumulative deviation amplitude exceeds 4. The deviation starting point that meets the conditions is confirmed as the pacing stimulation start point. Step c): Premature beat identification and sequence generation: Analyze the signal morphology characteristics between adjacent pacemakers, calculate the signal mean within the interval as a reference baseline, detect the deviation of the ECG signal from the baseline at ±5 sampling points near the pacemaker, and verify the temporal consistency in conjunction with the expected heartbeat interval; set the premature beat judgment condition as follows: the ECG signal deviation from the baseline exceeds 100, or the actual interval deviates from the expected interval by less than 20 sampling points. Based on the above analysis, generate the final pacing time sequence and premature beat marker sequence to complete the entire detection process.

[0013] In step S5, the specific process of detecting the R-wave peak value, calculating the S-V6RWP value, and the S-wave depth, based on the pacing time, is as follows: First, the ECG-V6 lead signal is preprocessed and its features are extracted. Based on the pacing time sequence, the signal is divided into individual heartbeat intervals. Mean filtering is applied to smooth the signal in each effective pacing interval, and cubic spline interpolation is used to interpolate the signal by a factor of 8. Next, R-wave peak detection is performed, and bandpass filtering is applied to the interpolated signal. The global maximum point in the first half of the interval is found as the initial R-wave position. Then, S-wave detection and parameter calculation are performed. In the signal segment after the R-wave, the S-wave is located by finding the negative peak. Constraints of peak height threshold 3, minimum distance 10, and width 5 are set to verify the candidate S-wave position. The actual minimum point is found within a range of 200 points around the candidate S-wave, and the difference between this point and the maximum value within the next 100 points is checked to see if it is greater than 2. The S-V6RWPT value is calculated as the time interval from the pacing moment S to the peak value of the R wave, multiplied by a time conversion factor to obtain the actual millisecond value; The S-wave depth is obtained by calculating the difference between the mean of the signal in the second half of the heartbeat interval and the S-wave trough value, and then multiplying it by the amplitude conversion coefficient of the ECG-V6 lead signal to obtain the actual voltage amplitude. At the same time, the baseline reference value and the starting height of the ECG-V6 lead signal for each heartbeat are recorded to provide a benchmark for subsequent visualization.

[0014] In step S6, the specific process of determining the baseline and locating the peak value in the EGM-LBB-I lead signal, and calculating the amplitude, is as follows: First, signal segmentation and baseline establishment are performed: based on the pacing time sequence, the EGM-LBB-I lead signal is segmented into individual heartbeat intervals, and the mean of the second half of the signal of each effective pacing interval is calculated as the baseline reference value for that heartbeat. Next, peak detection and verification are performed: a peak detection algorithm is applied to the first half of the interval, with constraints of a minimum distance of 10 and a width of 5, to eliminate potential noise peaks in the first 20 sampling points. The candidate point with the largest amplitude is selected from the detected peaks as the characteristic peak of the EGM-LBB-I lead signal. The significance of this peak is verified, requiring that the difference between the peak and the baseline reference value is greater than 10 pixel units. For peaks that pass verification, their absolute position in the global signal and the corresponding image height coordinates are recorded. Finally, the amplitude parameter is calculated: the peak amplitude is obtained by calculating the difference between the characteristic peak value and the baseline reference value, and the image height coordinates corresponding to the baseline are recorded to provide a spatial reference for the alignment and visualization of signals in each lead.

[0015] In step S7, the specific process of constructing the fusion early warning model based on the extracted parameters is as follows: Dynamic monitoring is achieved by calculating parameter changes between adjacent heartbeats. The system employs a two-level threshold warning mechanism to finely classify S-V6RWPT changes: a level one warning is triggered when the change is within the 5-10ms range, marked in yellow on the graphical interface and logged accordingly; a level two warning is triggered when two consecutive changes are within the 5-10ms range, marked in red and announced via voice; a level three warning is triggered when the change exceeds 10ms, marked in red and announced via voice. When S-wave deepening is detected simultaneously, the warning level is raised by one level. Special processing logic is set for premature beats, identified by state variables, so that even if parameter changes reach the warning threshold, a regular warning is not triggered. The EGM-LBB-I amplitude warning uses an absolute threshold judgment standard, combined with a preset amplitude conversion coefficient for standardization. A red warning is triggered when the standardized amplitude is below 6mV, recording a serious abnormality; a yellow warning is triggered and a voice warning is initiated when it is in the critical range of 6-7mV; and a value above 7mV is considered within the normal range.

[0016] The advantages of this invention compared to the prior art are: This invention uses a video capture card to acquire images from the MEA (Multi-Aspect Ratio). Through image quality assessment and motion estimation compensation algorithms, it corrects image tearing and misalignment caused by equipment aging, electromagnetic interference, or graphics card issues, ensuring complete signal extraction. Subsequently, color space conversion and color masking techniques are used to accurately separate the ECG-V6 lead signal, EGM-LBB lead signal, and EGM-LBB-I lead signal.

[0017] In terms of signal analysis, the system uses the EGM-LBB lead signal as a reference, and combines differential processing and the Teager energy operator to detect the pacing stimulation initiation point to achieve accurate positioning in noisy environments; it completes R-wave positioning, S-wave depth measurement and S-V6RWPT calculation in lead V6; and it supplements the myocardial injury current early warning parameters by measuring the amplitude of the EGM-LBB-I lead signal.

[0018] This invention establishes a multi-parameter fusion intelligent early warning system, integrating S-V6RWPT trends, S-wave depth evolution, and LBB-I amplitude levels. Through graded thresholds and logical judgments, it achieves multi-level early warnings, including yellow and red warnings, enabling early identification of electrode arrival in the left bundle branch region and timely warnings of complications such as perforation risk. It boasts strong compatibility, enabling convenient deployment through non-contact image acquisition and is applicable to different brands and models of MEA devices. The image compensation algorithm improves robustness to poor input quality, and the multi-parameter early warning model overcomes the limitations of single indicators, enhancing the reliability of auxiliary judgments. Attached Figure Description

[0019] Figure 1 This is the overall architecture diagram of the multi-parameter fusion early warning system of the present invention; Figure 2 This is a flowchart of the image compensation algorithm of the present invention; Figure 3 This is a schematic diagram of the electrocardiogram signal analysis process of the present invention; Figure 4 This is the logic diagram for classifying early warning levels in this invention. Detailed Implementation

[0020] The present invention will be further described in detail below with reference to the accompanying drawings: like Figure 1 As shown, this invention constructs a multi-parameter fusion early warning system for left bundle branch pacing based on continuous pacing mapping. The system acquires images from the MEA (Mean Interference Imager) using a video capture card. Through image quality assessment and motion estimation compensation algorithms, it corrects image tearing and misalignment caused by equipment aging, electromagnetic interference, or graphics card issues, ensuring complete signal extraction. Subsequently, using color space conversion and color masking techniques, it accurately separates signals from multiple leads, including ECG-V6 (V6 lead), EGM-LBB (LBB lead), and EGM-LBB-I (LBB-I lead). In signal analysis, the system uses the LBB lead as a benchmark, combining differential processing and the Teager energy operator to detect the pacing stimulation origin, achieving precise positioning in noisy environments. In the V6 lead, it completes R-wave positioning, S-wave depth measurement, and S-V6RWPT calculation. Finally, it supplements the myocardial injury current early warning parameters through LBB-I lead amplitude measurement.

[0021] This invention establishes a multi-parameter fusion intelligent early warning system, integrating S-V6RWPT trends, S-wave depth evolution, and LBB-I amplitude levels. Through graded thresholds and logical judgments, it achieves multi-level early warnings, including yellow and red warnings, enabling early identification of electrodes reaching the left bundle branch region and timely alerts. The system boasts strong compatibility, enabling convenient deployment through non-contact image acquisition. It is suitable for different brands and models of MEA devices, and the image compensation algorithm enhances robustness to poor-quality input.

[0022] A multi-parameter fusion early warning method for left bundle branch pacing based on continuous pacing mapping includes the following steps: Step S1: Extract ECG signal images frame by frame from the MEA video stream, perform ROI region cropping and grayscale processing to simplify subsequent calculations. This step constructs a complete ECG signal digital acquisition and preprocessing workflow, acquiring raw video stream data from the MEA through the video acquisition interface. This system supports multiple input source configurations, including real-time camera acquisition and local video file reading. In real-time acquisition mode, the system initializes the video capture device, configuring acquisition parameters including resolution, frame rate, and exposure settings to ensure a stable and continuous image sequence. The specific process is as follows: Fixed coordinate cropping technology is used to achieve region positioning, accurately locking the signal display area on the MEA screen through preset ROI parameters. The cropped image is first converted to grayscale, and the pixel grayscale value is calculated using a weighted average method to reduce data dimensionality while maintaining the contrast between the signal and the background. Subsequently, contrast enhancement and noise filtering are performed to improve the clarity of the signal trajectory. At the same time, an inter-frame temporal correlation model is established, and the accuracy of temporal analysis is ensured through timestamp recording and frame rate monitoring. The system maintains an image buffer to achieve smooth processing of continuous frames and data integrity verification.

[0023] Step S2: Image tearing is detected through inter-frame difference analysis, and compensation is performed using piecewise linear stitching and pixel interpolation to output a geometrically consistent ECG signal image. In consecutive frames, image tearing may occur due to acquisition or transmission issues, manifesting as broken or misaligned curves in the image. This step utilizes a multi-frame analysis-based image tearing detection and intelligent compensation algorithm, such as... Figure 2 As shown, the specific process is as follows: Step (1), continuous frame difference analysis: Establish an inter-frame signal continuity evaluation model through continuous frame difference analysis. The system maintains a double-frame buffer to store the current frame. And the previous frame The grayscale image is used to identify the effective signal region of the two frames: the non-zero state of each row of pixels is calculated, and a binary vector is generated. and Non-zero lines are marked as 1, otherwise as 0; the common valid line region of the two frames is obtained through a logical AND operation: Determine the common valid line range of the two frames. If the common valid line range is less than 60% of the image height or the starting line exceeds 40% of the image height, the image quality is considered poor and the current frame is skipped.

[0024] Step (2), Displacement Extraction: Within the effective row range, select the first and last two typical signal lines for displacement calculation. For each row, calculate the displacement d between the current frame and the corresponding row of the previous frame; the calculation process for displacement d is as follows: Step 1) Set This refers to a line of signal from the previous frame. This is the corresponding line signal for the current frame; Step 2) Find the set of pixel locations that satisfy the following conditions: , That is, the position where the current frame signal value is greater than the previous frame and the previous frame signal is zero; Step 3) If Y is not empty, then start from the first element of Y. Begin by finding the first satisfaction. The location, which is related to The difference is the displacement d. .

[0025] Step (3), Boundary Tear Location: By comparing the displacement of the first and last rows. and Determine if there is tearing in the image: Step a, if If so, it is considered that a rift exists, and the compensation process begins. To set a threshold (e.g., set to 5 pixels); Step b: To calculate the inter-line difference, define a horizontal window with a width defined as follows: and 2.5 times the maximum value; Step c: Within the horizontal window, calculate the difference between corresponding rows of two frames to determine the location of the tear boundary. The calculation process is as follows: in, The difference threshold, and It represents the starting position and width of the window area.

[0026] Step (4), Image Compensation: Based on the position of the torn boundary, divide the image into upper and lower halves, and adjust the displacement of the upper and lower halves accordingly. and The images are translated separately, and a zeroing region is set near the tear point to ensure a smooth transition. The translated upper and lower halves of the image and the zeroing region are then stitched together to obtain the compensated image.

[0027] Step S3: Apply color masking and morphological closing operations to separate the trajectories of ECG-V6 lead signal (V6 lead), EGM-LBB lead signal (LBB lead), and EGM-LBB-I lead signal (LBB-I lead). Generate the time-series signal sequences of each lead by scanning pixels. This step establishes a multi-lead signal separation method based on color space and morphological processing, the specific steps of which are as follows: First, the compensated RGB image obtained in step S2 is converted to the HSV color space. A lead color feature model is established using three independent dimensions: hue, saturation, and brightness. The system has a built-in standard lead color database, which contains the HSV threshold ranges of ECG-V6, EGM-LBB, and EGM-LBB-I lead signals. Initial binary masks for each lead signal are generated through color thresholding. Then, morphological operations are applied to optimize the segmentation results. A dilation-erosion closing operation is used to fill the small gaps in the signal trajectory and connect the broken areas caused by noise. At the same time, an opening operation is used to eliminate isolated noise points to improve the purity of the segmentation structure.

[0028] Signal trajectory reconstruction and optimization extracts continuous signal trajectories from each lead using column scanning technology. The system analyzes the binary mask column by column from left to right, recording the top position coordinates of the signal trajectory in each column. For columns with missing signals, a context-information-based interpolation filling algorithm is used. The reconstructed signal trajectory undergoes smoothing filtering to eliminate high-frequency noise introduced during acquisition and segmentation, while preserving the original morphological features of the ECG signal.

[0029] Step S4: Based on the signals from each lead, premature beats / pacing heartbeats are detected and the pacing stimulation initiation point is located using first-order difference, the Teager energy operator, and an adaptive dual-threshold method, outputting the pacing timing sequence. This step, through a multi-level detection and verification mechanism combined with signal processing and morphological analysis, achieves accurate identification of premature beats and precise location of the pacing stimulation initiation point under complex ECG conditions, providing a foundation for cardiac pacemaker function assessment and arrhythmia diagnosis. The specific process is as follows: Step a) Signal mutation point detection and feature enhancement: Calculate the first-order difference of the ECG signal sequence, identify the signal change trend, and apply the Teager energy operator (TEO) to enhance the transient characteristics of the signal. The calculation method is to subtract the product of the adjacent difference values ​​from the square of the current difference value, divide the result by 20 for amplitude limiting, and limit values ​​exceeding 200 to 200. Based on the TEO threshold being greater than 20, potential pacing stimulation initiation candidate locations are selected. At the same time, spatiotemporal correlation verification is performed by combining significant peak points with amplitudes exceeding 700 in the EGM-LBB-I lead signal. A matching tolerance range of 50 sampling points is set, and candidate point markers are established to identify reliable pacing points verified by multiple signals.

[0030] Step b) Fine screening and confirmation of pacemakers: For each candidate pacemaker, analyze the signal characteristics within the preceding 10 sampling point windows, including calculating the standard deviation of the original EGM-LBB lead signal and ECG-V6 lead signal to verify signal stationarity. If the standard deviation is greater than a set threshold, it is determined as a non-verification candidate point. For example, if the set threshold is 2, then a standard deviation ≤ 2 indicates a stationary signal segment; a standard deviation > 2 indicates a non-verification candidate point. For non-verification candidate points, a histogram statistical method is used to determine the signal baseline, calculate the difference between the signal and the baseline, and identify persistent significant deviation segments. The deviation judgment conditions are set as follows: the number of consecutive deviation sampling points is greater than 2 and the cumulative deviation amplitude exceeds 2, or it is close to the end of the window and the cumulative deviation amplitude exceeds 4. The deviation starting point that meets the conditions is confirmed as the pacing stimulation start point to ensure the accuracy of positioning.

[0031] Step c): Premature beat identification and sequence generation: Analyze the signal morphology characteristics between adjacent pacemakers, calculate the signal mean within the interval as a reference baseline, detect the deviation of the ECG signal from the baseline at ±5 sampling points near the pacemaker, and verify the temporal consistency in conjunction with the expected heartbeat interval. The premature beat determination criteria are set as follows: the ECG signal deviation from the baseline exceeds 100, or the actual interval deviates from the expected interval by less than 20 sampling points. Based on the above analysis, the final pacing time sequence and premature beat marker sequence are generated, completing the entire detection process.

[0032] Step S5: In the ECG-V6 lead signal, using the pacing time as a reference, detect the R-wave peak value, calculate the S-V6RWP value, and calculate the S-wave depth. This step uses the pacing stimulation signal (S) time as a reference to accurately detect the R-wave peak position in the ECG-V6 lead signal, calculate the S-V6RWPT, and quantify the S-wave depth of the ECG-V6 lead signal. The specific process is as follows: First, the ECG-V6 lead signal is preprocessed and its features are extracted. Based on the pacing time (S) sequence, the signal is divided into individual heartbeat intervals. Mean filtering (e.g., setting the kernel size to 3) is applied to smooth the signal for each effective pacing interval. Cubic spline interpolation is used to interpolate the signal by a factor of 8 to improve the temporal resolution for accurate feature point location. Next, R-wave peak detection is performed. The interpolated signal is bandpass filtered (0.5-45Hz) to remove baseline drift and high-frequency noise. The global maximum point is found in the first half of the interval as the initial R-wave position. Then, S-wave detection and parameter calculation are performed. In the signal segment after the R-wave, the S-wave is located by finding negative peaks. Constraints of peak height threshold 3, minimum distance 10, and width 5 are set to verify the candidate S-wave position. The actual minimum point is found within a 200-point range around the candidate S-wave, and the difference between this point and the maximum value within the next 100 points is checked to ensure the significance of the S-wave.

[0033] The T-V6RWPT value is calculated as the time interval from the pacing moment S to the peak value of the R wave, multiplied by a time conversion factor to obtain the actual millisecond value.

[0034] The S-wave depth is obtained by calculating the difference between the mean signal value of the latter half of the heartbeat interval and the S-wave trough value, and then multiplying it by the amplitude conversion factor of the ECG-V6 lead signal to obtain the actual voltage amplitude. Simultaneously, the baseline reference value and the initial height of the ECG-V6 lead signal for each heartbeat are recorded to provide a benchmark for subsequent visualization. The amplitude conversion factor mentioned above is a system setting parameter and can be manually set.

[0035] Step S6: In the EGM-LBB-I lead signal, determine the baseline and locate the peak value, and calculate the amplitude. This step involves establishing a signal baseline reference in the EGM-LBB-I lead signal, accurately locating the characteristic peak value, and calculating the corresponding amplitude parameters. The specific process is as follows: First, signal segmentation and baseline establishment are performed: Based on the pacing time sequence, the EGM-LBB-I lead signal is segmented into individual heartbeat intervals. The mean of the second half of the signal in each effective pacing interval is calculated as the baseline reference value for that heartbeat. This baseline reflects the DC component of the signal within that heartbeat cycle.

[0036] Next, peak detection and verification are performed: A peak detection algorithm is applied to the first half of the interval, with constraints of a minimum distance of 10 and a width of 5. Potential noise peaks within the first 20 sampling points are excluded. The candidate point with the largest amplitude is selected from the detected peaks as the characteristic peak of the EGM-LBB-I lead signal. This peak is then verified for significance, requiring the difference between the peak and the baseline reference value to be greater than 10 pixels to ensure the detected peak has physiological significance. For verified peaks, their absolute position in the global signal (pacing time plus relative offset) and the corresponding image height coordinates are recorded.

[0037] Finally, amplitude parameters are calculated: the peak amplitude is obtained by calculating the difference between the characteristic peak value and the baseline reference value, directly reflecting the amplitude characteristics of the EGM-LBB-I lead signal; at the same time, the image height coordinates corresponding to the baseline are recorded to provide a spatial reference for the alignment and visualization of each lead signal. All parameters are output in sequence form, maintaining a strict correspondence with the pacing time sequence to ensure timing consistency.

[0038] Step S7: Construct a fusion early warning model based on the extracted parameters, including S-V6RWP value, S-wave depth evolution analysis of ECG-V6 lead signal, and COI early warning of EGM-LBB-I lead signal, and provide visual early warning through a graphical interface and voice prompts. This step establishes a real-time dynamic monitoring mechanism by constructing a multi-parameter fusion early warning system of S-V6RWPT, S-wave depth, and LBB-I amplitude, to achieve intelligent identification and multimodal early warning of cardiac electrophysiological abnormalities, such as... Figure 4 As shown, the specific process is as follows: Dynamic monitoring is achieved by calculating parameter changes between adjacent heartbeats. The system employs a two-level threshold warning mechanism to finely classify S-V6RWPT changes: a level one warning is triggered when the change is within the 5-10ms range, marked in yellow on the graphical interface and logged accordingly; a level two warning is triggered when two consecutive changes are within the 5-10ms range, marked in red and announced via voice; a level three warning is triggered when the change exceeds 10ms, marked in red and announced via voice. When S-wave deepening is detected simultaneously, the warning level is raised by one level. Special processing logic is set for premature beats, identified by state variables, so that even if parameter changes reach the warning threshold, a regular warning is not triggered. The EGM-LBB-I amplitude warning uses an absolute threshold judgment standard, combined with a preset amplitude conversion coefficient for standardization. A red warning is triggered when the standardized amplitude is below 6mV, recording a serious abnormality; a yellow warning is triggered and a voice warning is initiated when it is in the critical range of 6-7mV; and a value above 7mV is considered within the normal range.

[0039] Through the aforementioned multi-parameter fusion early warning mechanism, this invention enables real-time monitoring and intelligent early warning of key electrophysiological parameters during pacing in the left bundle branch region of the heart, providing clinicians with intuitive and comprehensive decision-making references and effectively improving the safety and reliability of cardiac pacing therapy.

[0040] The above are merely preferred embodiments of the present invention and are not intended to limit the implementation methods and protection scope of the present invention. Those skilled in the art should recognize that any equivalent substitutions and obvious changes made based on the description and illustrations of the present invention should be included within the protection scope of the present invention.

Claims

1. A multi-parameter fusion early warning method for left bundle branch pacing based on continuous pacing mapping, characterized in that: Includes the following steps: Step S1: Extract ECG signal images frame by frame from the MEA video stream, and perform ROI region cropping and grayscale processing; Step S2: Detect image tearing through continuous inter-frame difference analysis, compensate using piecewise linear stitching and pixel interpolation, and output a geometrically consistent electrocardiogram signal image; Step S3: Apply color mask and morphological closing operation to separate the ECG-V6 lead signal, EGM-LBB lead signal and EGM-LBB-I lead signal trajectory, and generate the time signal sequence of each lead signal by scanning pixels; Step S4: Based on the signals from each lead, detect premature beats / pacing heartbeats and locate the pacing stimulation start point by using first-order difference, Teager energy operator and adaptive double threshold method, and output the pacing time sequence; Step S5: In the ECG-V6 lead signal, based on the pacing time, detect the R wave peak value, calculate the S-V6RWP value and S wave depth; Step S6: In the EGM-LBB-I lead signal, determine the baseline and locate the peak value, and calculate the amplitude; Step S7: Construct a fusion early warning model based on the extracted parameters, including S-V6RWP value, S-wave depth evolution analysis of ECG-V6 lead signal and COI early warning of EGM-LBB-I lead signal, and provide visual early warning through graphical interface and voice prompts.

2. The multi-parameter fusion early warning method for left bundle branch pacing based on continuous pacing mapping as described in claim 1, characterized in that: In step S1, fixed coordinate interception technology is used to achieve area positioning. The signal display area in the MEA screen is locked by preset ROI parameters, and ROI area interception is performed.

3. The multi-parameter fusion early warning method for left bundle branch pacing based on continuous pacing mapping as described in claim 1, characterized in that: In step S1, the specific process of grayscale processing is as follows: the weighted average method is used to calculate the pixel grayscale value, which reduces the data dimension while maintaining the contrast between the signal and the background.

4. The multi-parameter fusion early warning method for left bundle branch pacing based on continuous pacing mapping as described in claim 1, characterized in that: In step S2, the image tearing detection and compensation process includes the following steps: Step (1), consecutive frame difference analysis: The system maintains a double-frame buffer to store the current frame. And the previous frame The grayscale image is used to identify the effective signal region of the two frames: the non-zero state of each row of pixels is calculated, and a binary vector is generated. and Non-zero lines are marked as 1, otherwise as 0; The common valid line region of the two frames is obtained through a logical AND operation: Determine the common valid line range of two frames. If the common valid line range is less than 60% of the image height or the starting line exceeds 40% of the image height, the image quality is considered poor, and the current frame is skipped. Step (2), Displacement Extraction: Within the effective row range, select the first and last two typical signal lines for displacement calculation. For each row, calculate the displacement d between the current frame and the corresponding row of the previous frame; Step (3), Boundary Tear Location: By comparing the displacement of the first and last rows. and Determine if there is tearing in the image: Step a, if If so, it is considered that a rift exists, and the compensation process begins. To set a threshold; Step b: Define a horizontal window, with a width defined as [value to be filled in]. and 2.5 times the maximum value; Step c: Within the horizontal window, calculate the difference between corresponding rows of two frames to determine the location of the tear boundary. The calculation process is as follows: in, The difference threshold, and It refers to the starting position and width of the window area; Step (4), Image Compensation: Based on the position of the torn boundary, divide the image into upper and lower halves, and adjust the displacement of the upper and lower halves accordingly. and The images are translated separately, and a zeroing region is set near the tear point to ensure a smooth transition. The translated upper and lower halves of the image and the zeroing region are then stitched together to obtain the compensated image.

5. The multi-parameter fusion early warning method for left bundle branch pacing based on continuous pacing mapping as described in claim 4, characterized in that: In step (2), the calculation process for the displacement d is as follows: Step 1) Set This refers to a line of signal from the previous frame. This is the corresponding line signal for the current frame; Step 2) Find the set of pixel locations that satisfy the following conditions: , That is, the position where the current frame signal value is greater than the previous frame and the previous frame signal value was zero; Step 3) If Y is not empty, then start from the first element of Y. Begin by finding the first satisfaction. The location, which is related to The difference is the displacement d. 。 6. The multi-parameter fusion early warning method for left bundle branch pacing based on continuous pacing mapping as described in claim 1, characterized in that: In step S3, the process of separating the signals of each lead is as follows: First, the compensated RGB image obtained in step S2 is converted to the HSV color space. Then, a lead color feature model is established using three independent dimensions: hue, saturation, and brightness. The system has a built-in standard lead color database, which contains the HSV threshold ranges of ECG-V6 lead signals, EGM-LBB lead signals, and EGM-LBB-I lead signals. Initial binary masks for each lead signal are generated through color thresholding. Then, morphological operations are applied to optimize the segmentation results. A dilation-erosion closing operation is used to fill the small gaps in the signal trajectory and connect the broken areas caused by noise. At the same time, an opening operation is used to eliminate isolated noise points. The system analyzes the binary mask column by column from left to right, records the top position coordinates of the signal trajectory in each column, and uses a context-information-based interpolation filling algorithm for columns with missing signals. The reconstructed signal trajectory is then smoothed and filtered to eliminate high-frequency noise introduced during acquisition and segmentation, while preserving the original morphological features of the ECG signal.

7. The multi-parameter fusion early warning method for left bundle branch pacing based on continuous pacing mapping as described in claim 1, characterized in that: In step S4, the process of locating the pacing stimulation initiation point is as follows: Step a) Signal mutation point detection and feature enhancement: Calculate the first-order difference of the ECG signal sequence, and apply the Teager energy operator (TEO) to enhance the transient characteristics of the signal. The calculation method is to subtract the product of the adjacent difference values ​​from the square of the current difference value, and divide the result by 20 for amplitude limiting, and limit values ​​exceeding 200 to 200; select potential pacing stimulation initiation candidate positions based on the TEO threshold being greater than 20, and at the same time, combine the significant peak points with amplitudes exceeding 700 in the EGM-LBB-I lead signal for spatiotemporal correlation verification, set a matching tolerance range of 50 sampling points, and establish candidate point markers; Step b) Fine screening and confirmation of pacemakers: For each candidate pacemaker, analyze the signal characteristics within the preceding 10 sampling point windows, including calculating the standard deviation of the original EGM-LBB lead signal and ECG-V6 lead signal to verify signal stability. If the standard deviation is greater than a set threshold, it is determined as a non-verification candidate. For non-verification candidate, the histogram statistical method is used to determine the signal baseline, calculate the difference between the signal and the baseline, and identify persistent significant deviation segments. The deviation judgment criteria are set as follows: the number of consecutive deviation sampling points is greater than 2 and the cumulative deviation amplitude exceeds 2, or it is close to the end of the window and the cumulative deviation amplitude exceeds 4; the deviation starting point that meets the criteria is confirmed as the pacing stimulation starting point. Step c): Premature beat identification and sequence generation: Analyze the signal morphology characteristics between adjacent pacemakers, calculate the signal mean within the interval as a reference baseline, detect the deviation of the ECG signal from the baseline at ±5 sampling points near the pacemaker, and verify the temporal consistency in conjunction with the expected heartbeat interval; set the premature beat judgment condition as follows: the ECG signal deviation from the baseline exceeds 100, or the actual interval deviates from the expected interval by less than 20 sampling points. Based on the above analysis, generate the final pacing time sequence and premature beat marker sequence to complete the entire detection process.

8. The multi-parameter fusion early warning method for left bundle branch pacing based on continuous pacing mapping as described in claim 1, characterized in that: In step S5, the specific process of detecting the R-wave peak value, calculating the S-V6RWP value, and the S-wave depth, based on the pacing time, is as follows: First, the ECG-V6 lead signal is preprocessed and its features are extracted. Based on the pacing time sequence, the signal is divided into individual heartbeat intervals. Mean filtering is applied to smooth the signal in each effective pacing interval, and cubic spline interpolation is used to interpolate the signal by a factor of 8. Next, R-wave peak detection is performed, and bandpass filtering is applied to the interpolated signal. The global maximum point in the first half of the interval is found as the initial R-wave position. Then, S-wave detection and parameter calculation are performed. In the signal segment after the R-wave, the S-wave is located by finding the negative peak. Constraints of peak height threshold 3, minimum distance 10, and width 5 are set to verify the candidate S-wave position. The actual minimum point is found within a range of 200 points around the candidate S-wave, and the difference between this point and the maximum value within the next 100 points is checked to see if it is greater than 2. The U-V6RWPT value is calculated as the time interval from the pacing moment S to the peak value of the R wave, multiplied by a time conversion factor to obtain the actual millisecond value; The S-wave depth is obtained by calculating the difference between the mean of the signal in the second half of the heartbeat interval and the S-wave trough value, and then multiplying it by the amplitude conversion coefficient of the ECG-V6 lead signal to obtain the actual voltage amplitude. At the same time, the baseline reference value and the starting height of the ECG-V6 lead signal for each heartbeat are recorded to provide a benchmark for subsequent visualization.

9. The multi-parameter fusion early warning method for left bundle branch pacing based on continuous pacing mapping as described in claim 1, characterized in that: In step S6, the specific process of determining the baseline and locating the peak value in the EGM-LBB-I lead signal, and calculating the amplitude, is as follows: First, signal segmentation and baseline establishment are performed: based on the pacing time sequence, the EGM-LBB-I lead signal is segmented into individual heartbeat intervals, and the mean of the second half of the signal of each effective pacing interval is calculated as the baseline reference value for that heartbeat. Next, peak detection and verification are performed: a peak detection algorithm is applied to the first half of the interval, with constraints of a minimum distance of 10 and a width of 5, to eliminate potential noise peaks in the first 20 sampling points. The candidate point with the largest amplitude is selected from the detected peaks as the characteristic peak of the EGM-LBB-I lead signal. The significance of this peak is verified, requiring that the difference between the peak and the baseline reference value is greater than 10 pixel units. For peaks that pass verification, their absolute position in the global signal and the corresponding image height coordinates are recorded. Finally, the amplitude parameter is calculated: the peak amplitude is obtained by calculating the difference between the characteristic peak value and the baseline reference value, and the image height coordinates corresponding to the baseline are recorded to provide a spatial reference for the alignment and visualization of signals in each lead.

10. The multi-parameter fusion early warning method for left bundle branch pacing based on continuous pacing mapping according to claim 1, characterized in that: In step S7, the specific process of constructing the fusion early warning model based on the extracted parameters is as follows: Dynamic monitoring is achieved by calculating parameter changes between adjacent heartbeats. The system employs a two-level threshold warning mechanism to finely classify S-V6RWPT changes: a level one warning is triggered when the change is within the 5-10ms range, marked in yellow on the graphical interface and logged accordingly; a level two warning is triggered when two consecutive changes are within the 5-10ms range, marked in red and announced via voice; a level three warning is triggered when the change exceeds 10ms, marked in red and announced via voice. When S-wave deepening is detected simultaneously, the warning level is raised by one level. Special processing logic is set for premature beats, identified by state variables, so that even if parameter changes reach the warning threshold, a regular warning is not triggered. The EGM-LBB-I amplitude warning uses an absolute threshold judgment standard, combined with a preset amplitude conversion coefficient for standardization. A red warning is triggered when the standardized amplitude is below 6mV, recording a serious abnormality; a yellow warning is triggered and a voice warning is initiated when it is in the critical range of 6-7mV; and a value above 7mV is considered within the normal range.

Images