Atrial fibrillation radiofrequency ablation postoperative recurrence prediction method based on multi-modal analysis

Abstract

The invention relates to an atrial fibrillation radiofrequency ablation postoperative recurrence prediction method based on multi-modal analysis. Comprising the following steps: dynamically capturing the form and spatial distribution of a low-voltage region by using a three-dimensional electroanatomical map; introducing a regional complexity quantitative index, and analyzing the non-linear degree of the boundary of the low-voltage region; grading the voltage intensity gradient in the low-voltage region, and quantifying the sensitivity to postoperative recurrence; establishing a composite index model of the area ratio of the low-voltage region and the fragmentation potential intensity; based on spatial distribution and voltage features, performing high, medium and low recurrence risk classification on the low-voltage region, and outputting a classification heat map; analyzing the migration path of the fragmentation potential in the three-dimensional space in the operation, and extracting the migration speed, direction and frequency change; discovering a fragmentation potential concentration trend of the high recurrence sensitive area by using a trajectory clustering algorithm; and extracting spectrum characteristics of the fragmentation potential signal based on wavelet transform, and identifying abnormal distribution of the high-frequency fragmentation potential.

Description

Technical Field

[0001] The present invention relates to a method for predicting the recurrence of atrial fibrillation after radiofrequency ablation, and in particular to a method for predicting the recurrence of atrial fibrillation after radiofrequency ablation based on multimodal analysis. Background Art

[0002] Among the current methods for predicting recurrence after radiofrequency ablation of atrial fibrillation, the late recurrence risk prediction device and system after atrial fibrillation ablation proposed in Chinese invention patent 202310805328X are representative. Although the risk of late recurrence can be preliminarily evaluated through postoperative ECG data, there are still many deficiencies and disadvantages in terms of prediction accuracy, multi-dimensional feature integration, dynamic analysis capabilities, and support for personalized guidance. The following is a detailed description of its core methods and implementation logic. The main feature of this method is to classify and predict the risk of late recurrence based on whether ultra-early recurrence occurs, the relationship between atrial fibrillation load and classification threshold by obtaining the patient's postoperative ECG data. A significant shortcoming of this method is that it relies on single-modality ECG data, especially the simple calculation of the presence or absence of ultra-early recurrence (atrial fibrillation within 7 days after surgery) and the atrial fibrillation load, ignoring deeper physiological factors such as the spatial distribution of atrial fibrillation, changes in electrophysiological characteristics, and fibrosis expansion. Postoperative recurrence is the result of the combined action of multiple factors and multiple stages. Simply relying on the statistical analysis of postoperative ECG data cannot fully capture the complex dynamic evolution of postoperative recurrence, so the accuracy and applicability of the prediction are limited.

[0003] Secondly, this method has a relatively simple definition of ultra-early recurrence, focusing only on whether atrial fibrillation occurs within 7 days after surgery, without conducting an in-depth analysis of the specific characteristics of atrial fibrillation (such as migration path, frequency change, and distribution of fragmented potentials). For example, the dynamic behavior and spectral characteristics of fragmented potentials in different regions may significantly affect the risk of postoperative recurrence, but this method fails to combine spatial characteristics with time series for joint modeling. Simply judging "whether there is atrial fibrillation" may lead to missed diagnosis of some high-risk patients, especially the heterogeneity of atrial fibrillation characteristics between early and late postoperative periods cannot be fully identified. In addition, the calculation of atrial fibrillation load in this method is limited to the proportion of total atrial fibrillation duration within 7 days after surgery, and this is used as the main determinant of late recurrence risk. This threshold-based classification method fails to effectively capture the dynamic trend of atrial fibrillation load changes. For example, the postoperative atrial fibrillation load may show nonlinear fluctuations in different time periods, and this fluctuation pattern may better reflect the risk of recurrence than the static load value. However, this method does not take into account the dynamic evolution of the load over time, resulting in the inability of the prediction model to capture the key time nodes of the risk of recurrence, thereby reducing the accuracy of risk assessment. At the same time, this method mainly relies on early postoperative ECG data for the prediction of late recurrence risk 90 days after surgery, and fails to include postoperative follow-up imaging data (such as cardiac MRI or CT) and preoperative characteristics (such as fibrosis degree and low-voltage area distribution) in the analysis. Postoperative follow-up imaging data can provide important information on atrial fibrosis extension, low-voltage area changes, and spatial characteristics, which are important predictive factors for late recurrence. However, due to the lack of imaging support, this method has obvious limitations in modeling spatial dynamic characteristics.

[0004] In addition, this method fails to fully consider the individual differences of patients. The risk of recurrence of atrial fibrillation is not only related to postoperative ECG data, but also affected by multiple factors such as the patient's underlying diseases (such as hypertension, diabetes), the degree of atrial remodeling, and postoperative management. The current classification method divides all patients' ultra-early recurrence and atrial fibrillation burden according to fixed thresholds, and does not optimize the prediction model for different patient groups (such as patients with chronic atrial fibrillation and paroxysmal atrial fibrillation), resulting in a lack of personalization and accuracy in the prediction results.

[0005] Finally, this method lacks the ability to intuitively present prediction results and dynamic risk assessment. Simple classification results (high or low risk of recurrence) can hardly reflect the changing trend of patients' postoperative risk over time, nor can they intuitively show the contribution of high-risk areas or specific features to the risk. In contrast, generating a dynamic recurrence probability curve or recurrence sensitivity heat map can more intuitively help clinicians determine the key areas for postoperative follow-up and intervention, which is what this method lacks. Summary of the invention

[0006] The purpose of the present invention is to provide a method for predicting recurrence of atrial fibrillation after radiofrequency ablation based on multimodal analysis, thereby solving some of the drawbacks and shortcomings pointed out in the background technology.

[0007] The present invention solves the above-mentioned technical problems by adopting the following technical solution, which includes the following steps:

[0008] S1. Dynamic mapping and classification of intraoperative low voltage area characteristics:

[0009] S1.1. Use three-dimensional electroanatomical images to dynamically capture the morphology and spatial distribution of low-voltage areas; introduce regional complexity quantitative indicators to analyze the nonlinearity of low-voltage area boundaries;

[0010] S1.2. Classify the voltage intensity gradient within the low-voltage area and quantify the sensitivity to postoperative recurrence; establish a composite index model of low-voltage area proportion and fragmentation potential density;

[0011] S1.3. Based on spatial distribution and voltage characteristics, classify low voltage areas into high, medium, and low recurrence risks and output classification heat maps;

[0012] S2. Spatial migration pattern and spectrum characteristics analysis of fragmentation potential:

[0013] S2.1. Analyze the migration path of intraoperative fragmentation potential in three-dimensional space, extract the migration speed, direction and frequency changes; use trajectory clustering algorithm to find the concentration trend of fragmentation potential in high recurrence sensitive areas;

[0014] S2.2. Extract the spectral characteristics of the fragmentation potential signal based on wavelet transform and identify the abnormal distribution of high-frequency fragmentation potential; design a frequency and space coupling model to integrate the temporal dynamics and spatial distribution of the fragmentation potential;

[0015] S2.3. Use feature selection algorithms including LASSO to screen key features of fractionated potentials associated with postoperative recurrence risk;

[0016] S3. 3D reconstruction and erosion modeling of atrial fibrosis areas:

[0017] S3.1. Reconstruct the three-dimensional geometry of the fibrotic region based on postoperative cardiac MRI or CT data; extract the surface area, volume and distribution density of the fibrotic region;

[0018] S3.2. Analyze the erosion trajectory of the fibrotic area on the adjacent low-voltage area and healthy tissue; introduce the erosion rate index to quantify the degree of damage of fibrosis to the electrophysiological characteristics of the atrial heart;

[0019] S3.3. Construct a recurrence sensitivity distribution map of the fibrosis area and overlay it with the coupling area of ​​the low voltage area for analysis;

[0020] S4. Prediction of postoperative recurrence risk based on time series modeling:

[0021] S4.1. Extract time series features including voltage dynamic changes and fibrosis extension from intraoperative electrophysiological data and postoperative imaging follow-up data; use long short-term memory network (LSTM) to capture the dynamic change pattern of time series;

[0022] S4.2. Combine historical data to predict the time window for postoperative recurrence; output dynamic recurrence probability curve to provide time-sensitive risk assessment;

[0023] S5. Recurrence prediction model based on multimodal feature fusion:

[0024] S5.1. Design a multimodal feature alignment network to unify intraoperative electrophysiological features, postoperative imaging features, and clinical features; use an adaptive weight allocation algorithm to optimize the feature alignment effect;

[0025] S5.2. A deep graph neural network (GNN) is used to model the interaction between low voltage areas, fragmentation potentials, and fibrosis areas. The model outputs personalized recurrence probability and generates a sensitivity map of high-risk areas.

[0026] Furthermore, the spatial migration pattern and spectrum characteristic analysis method of the fragmentation potential includes:

[0027] Through the spatial displacement between adjacent time points, the instantaneous speed is calculated to identify the high dynamic area; based on the three-dimensional spatial vector change, the directional characteristic data set of potential movement is constructed; through the change of activation frequency within a fixed time window, the high active area and the stable area are distinguished, and the migration speed is calculated by the following formula:

[0028]

[0029] in:

[0030] v i is the instantaneous speed of the i-th fragmentation potential, describing the i to i+1 The movement rate between i ,y i ,z i is the fragmentation potential at time t i The three-dimensional space coordinates at the time; i+1 ,y i+1 ,z i+1 is the fragmentation potential at time t i+1 The three-dimensional space coordinates at time (t i+1 t i ) is the time difference between two time points; by calculating the spatial distance between two adjacent coordinate points and combining the time difference, the movement speed of the potential is quantified;

[0031] The direction extraction is expressed as a motion vector:

[0032]

[0033] in:

[0034] is the moving direction vector of the fragmentation potential, indicating that at time point t i to i+1 The direction of movement; x i+1 x i ,y i+1 y i ,z i+1 z i is the displacement corresponding to the three-dimensional space; it is used for directional cluster analysis;

[0035] The frequency change is calculated by the activation frequency fluctuation within a fixed time window:

[0036]

[0037] in:

[0038] f k is the activation frequency of the fragmentation potential in the time window k; N k is the total number of fragmentation potentials in time window k; δ i Is a mark for whether the fragmentation potential is activated, 0 means inactive, 1 means active; T k is the length of the time window.

[0039] Furthermore, the spatial migration pattern and spectrum characteristic analysis method of the fragmentation potential includes:

[0040] The path of each fragmentation potential is abstracted into a three-dimensional trajectory curve, including the time dimension. The dynamic time warping (DTW) algorithm is used to calculate the similarity between trajectories, highlighting the comprehensive characteristics of migration speed, direction, and frequency. Based on the clustering results, the areas with high density distribution of fragmentation potentials are marked to quantify the potential contribution to postoperative recurrence. The trajectory similarity measurement formula is:

[0041]

[0042] in:

[0043] For the trajectory and The dynamic similarity between i,k ,y i,k ,z i,k For the trajectory The three-dimensional space coordinates at the kth time point; xj,k ,y j,k ,z j,k For the trajectory The three-dimensional spatial coordinates at the kth time point; K is the total number of time points included in the trajectory.

[0044] Furthermore, the spatial migration pattern and spectrum characteristic analysis method of the fragmentation potential includes:

[0045] The sensitive areas in the clustering results were modeled to analyze the concentration trend and directional distribution of the fragmentation potential. The recurrence sensitivity distribution map was constructed based on the density and migration intensity of the fragmentation potential in the sensitive area. The characteristic report of the recurrence sensitivity area was generated based on the heat map results, and the migration intensity distribution model was calculated using the following formula:

[0046]

[0047] in:

[0048] M(x,y,z) is the migration intensity distribution of the target point (x,y,z); is the three-dimensional coordinate of the target point; is the three-dimensional coordinate of the trajectory point; σ is the adjustment parameter of the influence range, which controls the diffusion degree of the heat map distribution; t1, t2 are the time intervals of analysis; t i+1 t i is the time difference between adjacent time points.

[0049] Furthermore, the three-dimensional reconstruction and erosion modeling method of the atrial fibrosis region includes:

[0050] Based on cardiac MRI or CT images, the three-dimensional morphology of the fibrosis area was reconstructed and the distribution characteristics of the boundary points were extracted. The image data at multiple time points including 3 months, 6 months and 1 year after surgery were analyzed to record the expansion rate of the volume of the fibrosis area and the boundary migration trajectory. The contact points and expansion paths with the adjacent low-voltage areas and healthy tissues were identified by tracking the motion trajectory of the fibrosis boundary points. The volume expansion rate of the fibrosis area was quantified by the following formula:

[0051]

[0052] in:

[0053] R v (t) is the volume expansion rate of the fibrosis area at time t, indicating the change amplitude of the volume of the fibrosis area per unit time; is the volume of the fibrosis area at time t, which is defined as the set of spatial points inside the area; ρ(x, y, z) is the fibrosis density distribution function, which indicates the density of fibrosis at a three-dimensional point (x, y, z); dV is a small volume element, which is used to integrate the overall volume of the fibrosis area; Δt is the time interval, which is used to quantify the rate of change of the fibrosis volume per unit time.

[0054] Furthermore, the three-dimensional reconstruction and erosion modeling method of the atrial fibrosis region includes:

[0055] The overlap between the fibrosis area and the low voltage area in three-dimensional space was compared to quantify the erosion behavior of erosion depth, area and density changes. A dynamic coupling model was established by combining the electrophysiological characteristics of the fibrosis area and the low voltage area, including voltage amplitude and signal fluctuation, to evaluate the degree of damage to the function of the low voltage area caused by fibrosis extension. The erosion rate of the fibrosis area and the low voltage area was described by the following formula:

[0056]

[0057] in:

[0058] I r (t) is the erosion rate at time t, indicating the degree of damage of the fiberized area to the boundary of the low voltage area; is the interface between the fibrosis area and the low voltage area, which is the overlapping boundary of the two areas in three-dimensional space; κ(x, y, z) is the erosion depth function, which represents the erosion depth at the interface point (x, y, z); n(x, y, z) is the normal component function, which represents the distribution characteristics of the erosion direction on the interface, and is used to distinguish inward or outward erosion behavior; dS is a small area element on the interface, which is used to integrate and calculate the erosion rate of the entire interface.

[0059] Furthermore, the erosion behavior classification includes:

[0060] P1, edge erosion: the boundary of the fibrosis area gradually approaches the low voltage area, but does not completely cover it;

[0061] P2, full coverage erosion: the fibrosis area completely covers the low voltage area and further expands;

[0062] P3. Healthy tissue expansion: Fibrosis erodes into healthy tissue areas, resulting in secondary generation of low voltage areas.

[0063] 8. The method for predicting recurrence of atrial fibrillation after radiofrequency ablation based on multimodal analysis according to claim 6, characterized in that the three-dimensional reconstruction and erosion modeling method of the atrial fibrosis area comprises:

[0064] Quantify the effect of fibrosis expansion on the reduction of low voltage areas or the transformation of healthy tissue into low voltage areas; record the dynamic changes in the rate of fibrosis expansion, and identify the rapid stage and relatively stable stage of expansion; based on the expansion speed and erosion depth, mark the areas with high erosion risk as high-risk areas; the erosion rate index is defined as:

[0065]

[0066] in:

[0067] E(t) is the erosion rate index at time t, reflecting the overall degree of damage to the low voltage area caused by fiber expansion; T is the length of the analysis period; is the area of ​​the fiberization erosion region at time t, indicating the degree of influence of the low voltage area; γ(x, y, z) is the area change weight function, describing the contribution of a certain point in the fiberization region to the overall erosion; is the area of ​​the initial low-voltage zone, which is used as a normalized benchmark to represent the area of ​​the low-voltage zone before surgery; dA is a small area element, which is used to integrally calculate the impact of the fibrosis area on the entire low-voltage zone.

[0068] The method for predicting recurrence of atrial fibrillation after radiofrequency ablation based on multimodal analysis of the present invention has the following beneficial effects:

[0069] By integrating intraoperative electrophysiological characteristics, postoperative imaging follow-up data and clinical characteristics, the present invention realizes the fusion and dynamic modeling of multimodal data, and can accurately quantify the risk of recurrence after radiofrequency ablation of atrial fibrillation. In particular, the spatial distribution, dynamic behavior and interaction of low voltage areas, fragmented potentials and fibrosis areas are analyzed in a fine-grained manner, providing a scientific basis for personalized risk prediction.

[0070] The introduction of the migration speed, directionality and spectral characteristic analysis method of fragmented potentials, combined with dynamic time warping (DTW) and three-dimensional trajectory clustering algorithm, can accurately capture the dynamic behavior of fragmented potentials in the atrium; at the same time, through the modeling of the volume expansion rate, boundary migration trajectory and erosion depth of the fibrotic area, the erosion effect of fibrosis on low-voltage areas and healthy tissues is comprehensively analyzed, laying the foundation for the accurate marking of high-risk areas for postoperative recurrence.

[0071] In view of the interactive relationship between fibrosis areas and low-voltage areas, the present invention proposes a dynamic analysis method based on erosion rate and erosion depth, classifies fibrosis expansion behavior into three types: edge erosion, full coverage erosion, and healthy tissue expansion, and quantifies its contribution to postoperative recurrence through the migration intensity distribution model and erosion rate index, which can scientifically mark high-risk erosion areas and provide guidance for precise postoperative intervention. The use of time series modeling algorithms such as long short-term memory networks (LSTM) can capture the dynamic change patterns of features at multiple time points during and after surgery, and combine deep graph neural networks (GNN) to model the interactive relationship between low-voltage areas, fragmentation potentials, and fibrosis areas, to achieve accurate prediction of personalized recurrence probability. The output of dynamic recurrence probability curves and recurrence sensitivity heat maps provides clinicians with an intuitive and comprehensive risk assessment tool. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] Figure 1 This is a flow chart of the method for predicting recurrence of atrial fibrillation after radiofrequency ablation based on multimodal analysis of the present invention.

[0073] Figure 2 It is a flow chart of the spatial migration pattern and spectrum characteristic analysis method of the fragmentation potential of the present invention.

[0074] Figure 3 This is a flow chart of the three-dimensional reconstruction and erosion modeling method of the atrial fibrosis area of ​​the present invention.

[0075] Figure 4 This is a three-dimensional electroanatomical map and a low voltage area measurement map of Example 3 of the present invention.

[0076] Figure 5 This is a diagram of the dynamic distribution and migration path of the fragmentation potential of Example 3 of the present invention. DETAILED DESCRIPTION

[0077] The specific implementation modes of the present invention will be described in detail below in conjunction with the accompanying drawings.

[0078] The key technologies of the method for predicting recurrence of atrial fibrillation after radiofrequency ablation based on multimodal analysis include dynamic mapping and classification of intraoperative low-voltage zone characteristics. The specific methods are as follows: First, the morphological and spatial distribution characteristics of the low-voltage zone are captured in real time through three-dimensional electroanatomical maps, and its boundary changes and internal structures at different time points are recorded; in order to more comprehensively characterize the complexity of the low-voltage zone, a regional complexity quantification index is innovatively introduced. This index provides accurate regional complexity quantification results by analyzing the nonlinear degree of the low-voltage zone boundary, such as curvature changes and boundary irregularities; secondly, the voltage intensity gradient inside the low-voltage zone is graded, and different voltage gradient ranges are mapped to the region. The sensitivity levels of the atria are divided into high, medium and low, and the contribution of each area to the risk of postoperative recurrence is clarified; on this basis, a composite indicator model is established by combining the two important characteristics of the low-voltage area proportion and the fragmented potential density. This model can combine the spatial distribution characteristics of the low-voltage area with the level of electrophysiological activity to enhance the quantification ability of recurrence sensitivity; finally, based on the above feature analysis results, a low-voltage area recurrence risk classification model is constructed, and the low-voltage area is classified into high, medium and low recurrence risks through spatial distribution and voltage characteristics, and a risk classification heat map is output. The heat map intuitively shows the recurrence risk levels of different areas in the atrium, providing an intuitive reference for postoperative intervention.

[0079] By real-time monitoring of the three-dimensional spatial position of the intraoperative fragmented potential, recording its coordinate changes at different times, the migration speed, direction and frequency change characteristics of the fragmented potential are extracted. The speed reflects the dynamic activity of the potential, the direction represents its migration trend, and the frequency change captures its temporal activity characteristics. Then, based on the spatial migration path of the fragmented potential, a trajectory clustering algorithm (such as the dynamic time warping algorithm DTW or the density-based DBSCAN) is used to perform similarity analysis and classification on the trajectory of the fragmented potential, and the concentration trend of the fragmented potential in the high recurrence sensitive area is found, and these areas are marked. Furthermore, the wavelet transform is used to perform spectral analysis on the fragmented potential signal, extract its frequency characteristics, identify the abnormal distribution characteristics of high-frequency fragmented potentials, and clarify its active area in space.

[0080] To further improve the accuracy of the analysis, a frequency and space coupling model was designed. By integrating the temporal dynamics and spatial distribution of fragmented potentials, a unified representation was established to describe the complex behavior of fragmented potential activity in time and space. Finally, feature selection algorithms including LASSO (least absolute shrinkage and selection operator) were used to screen out key features of fragmented potentials that were significantly correlated with the risk of postoperative recurrence from multidimensional features such as frequency, space, and time, forming a concise and effective feature set.

[0081] Based on postoperative cardiac MRI or CT data, the three-dimensional geometric morphology of the fibrosis area is reconstructed using three-dimensional image reconstruction technology to accurately obtain the spatial distribution of the fibrosis area in the atrium and its boundary characteristics; based on the reconstruction, the key geometric parameters of the fibrosis area are extracted, including surface area, volume and distribution density, where the surface area reflects the complexity of the fibrosis boundary, the volume represents the cumulative degree of fibrosis, and the distribution density measures the spatial expansion intensity of fibrosis; then, by dynamically tracking the time series data of the fibrosis area, the erosion trajectory of the adjacent low-voltage area and healthy tissue is analyzed, the erosion path, erosion depth and expansion speed of the fibrosis boundary are specifically identified, and the key influencing areas of fibrosis expansion are revealed;

[0082] In order to quantitatively describe the damage of fibrosis to the electrophysiological characteristics of the atrium, an erosion rate index was introduced. This index quantified the intensity of the erosion behavior by combining the temporal and spatial characteristics of fibrosis expansion, and evaluated the degree of damage to the boundary of the low-voltage zone and healthy atrial tissue. Finally, based on the spatial distribution and erosion characteristics of the fibrosis area, a recurrence sensitivity distribution map of the fibrosis area was constructed. This distribution map intuitively displayed the sensitivity of the fibrosis area to the risk of postoperative recurrence. At the same time, the sensitivity distribution of the fibrosis area was superimposed and analyzed with the spatial coupling area of ​​the low-voltage area to comprehensively evaluate the interaction between the two, providing more comprehensive spatial information support and quantitative basis for the prediction of postoperative recurrence risk.

[0083] Time series features are extracted from intraoperative electrophysiological data and postoperative imaging follow-up data. These features include the dynamic changes of intraoperative low-voltage areas, the temporal trend of voltage intensity, and the expansion trajectory and erosion behavior of postoperative fibrosis areas. The extraction of time series features emphasizes the integration of multimodal data, combining the electrophysiological features captured in real time during surgery with the spatial dynamic changes of fibrosis areas in postoperative images, providing comprehensive time series input for recurrence risk modeling.

[0084] Next, a long short-term memory network (LSTM) was used to model the time series data. LSTM captures the dynamic change patterns of features in short-term and long-term time periods through its gating mechanism, and is particularly suitable for processing the complex temporal relationship between fibrosis expansion and voltage changes in postoperative follow-up data. By training the LSTM model, multi-time point data are mapped into a time pattern for recurrence risk prediction, and the time window for recurrence is accurately predicted by combining the patient's historical data and postoperative follow-up data. Based on the time window prediction, a dynamic recurrence probability curve is output, which shows the change in the patient's recurrence probability in different time periods after surgery through a time axis, providing clinicians with a time-sensitive basis for recurrence risk assessment. The output of the dynamic recurrence probability curve intuitively displays the evolution trend of postoperative risks, helping to determine the key time nodes for postoperative management, such as review cycles, intensive intervention opportunities, and long-term risk management plans, thereby improving the accuracy and timeliness of postoperative management.

[0085] A multimodal feature alignment network was designed to unify intraoperative electrophysiological features, postoperative imaging features and clinical features of patients into the same feature space. Electrophysiological features include the distribution of low voltage areas and the dynamic behavior of fragmented potentials. Imaging features include the volume, distribution and dynamic expansion trajectory of fibrosis areas. Clinical features include the age, gender and postoperative recovery of patients. Through the alignment of multimodal features, information complementarity and unification of feature dimensions between different modal features were achieved. To improve the accuracy and effectiveness of feature fusion, an adaptive weight allocation algorithm was used to assign optimal weights to different modal features. The weight distribution was dynamically adjusted according to the importance of the features to the prediction target, so that high-contribution features occupied a higher proportion in the model, thereby optimizing the feature alignment effect. Then, a deep graph neural network (GNN) was used to model the interactive relationship between low voltage areas, fragmented potentials and fibrosis areas. These areas and features were constructed into a graph structure. Nodes represent the feature information of each area, and edges represent the relationship between areas. GNN was used to capture the spatial and attribute interactions between features, revealing the joint impact of electrophysiological activity and fibrosis expansion on postoperative recurrence.

[0086] In the model output layer, based on the fusion of multimodal features and deep modeling of GNN, personalized recurrence probability is generated to provide a quantitative assessment of recurrence risk for each patient. At the same time, the model generates sensitivity maps of high-risk areas to intuitively mark areas with significant recurrence risk in the atrium. The sensitivity maps combine the spatiotemporal interaction characteristics of low-voltage areas, fragmented potentials, and fibrosis areas, providing intuitive guidance for postoperative recurrence management and precise intervention in high-risk areas, thereby achieving a comprehensive solution from multimodal feature analysis to recurrence risk prediction.

[0087] Embodiment 1:

[0088] A 65-year-old patient, Mr. Zhang, underwent radiofrequency ablation for atrial fibrillation. During the postoperative follow-up, the intraoperative electrophysiological data were analyzed using a multimodal analysis-based method for predicting recurrence after radiofrequency ablation of atrial fibrillation, focusing on the spatial migration pattern and spectral characteristics of fragmented potentials to assess the risk of postoperative recurrence.

[0089] In the intraoperative three-dimensional electroanatomical map, the trajectory of Mr. Zhang's fragmented potential in three-dimensional space is recorded. The three-dimensional coordinate data of the fragmented potential in a certain period of time are as follows:

[0090] Time t1 = 0.2s, position (x1, y1, z1) = (5, 3, 1);

[0091] Time t2 = 0.4s, position (x2, y2, z2) = (6, 4, 2).

[0092] According to the formula:

[0093]

[0094] Substitute into the calculation:

[0095]

[0096] The results show that the instantaneous speed of the fragmentation potential in this time interval is 8.66 cm / s, indicating that it is highly active. Based on the above time point data, the movement direction vector of the fragmentation potential is calculated:

[0097]

[0098] Substitute into the calculation:

[0099]

[0100] The movement direction vector is (1,1,1), indicating that the fragmentation potential moves along the diagonal direction in three-dimensional space. This information can be used for further directional clustering analysis.

[0101] Calculation of fragmentation potential frequency change:

[0102] In a fixed time window T k = 1s, record the activation of the fragmentation potential, assuming that at T k = 1s, the fragmentation potential activated marker δ i =[1,0,1,1,0], total N k = 5 observations, the number of activations is 3. According to the formula:

[0103]

[0104] Substitute into the calculation:

[0105]

[0106] The results showed that the activation frequency within this time window was 3 Hz, which belongs to the higher frequency area, suggesting that it is more sensitive to the risk of recurrence.

[0107] Combined with the above calculation results, Mr. Zhang's fragmented potentials showed high dynamic activity during the operation (instantaneous speed 8.66cm / s), regular direction (vector direction (1,1,1)), and high activation frequency (3Hz) within a fixed time window. These characteristics suggest that the fragmented potential may be located in an area with a higher risk of recurrence. Furthermore, through the trajectory clustering algorithm, this potential can be clustered with other similar highly active potentials to identify highly recurrence-sensitive areas in Mr. Zhang's atrium, providing a basis for precise postoperative management and intervention.

[0108] Coefficient value range:

[0109] 1. Coordinate units: x, y, z are in centimeters, ranging from 0 ≤ x, y, z ≤ 20 cm;

[0110] 2. Time interval: Δt = t i+1 -t i , generally take 0.1s≤Δt≤1s;

[0111] 3. Activation marker δ i : 0 means not activated, 1 means activated;

[0112] 4. Time window T k : Generally choose 0.5s≤T k ≤2s.

[0113] Through example analysis and computational verification, this method is feasible and practical, and provides quantitative support for postoperative recurrence risk assessment, demonstrating its important application value in actual clinical practice.

[0114] Based on the intraoperative electrophysiological data analysis, this example further conducts an in-depth analysis of the spatial migration pattern and spectral characteristics of the fragmented potentials, and combines three-dimensional trajectory curve modeling and trajectory similarity clustering to assess the risk of postoperative recurrence. and The trajectory data of Contains 5 time points, and the three-dimensional space coordinates are (x i,k ,y i,k ,z i,k ) = {(1,2,1),(2,3,2),(3,3,2),(4,4,3),(5,5,4)}; trajectory Contains the coordinates of the same time point, respectively (x j,k ,y j,k ,z j,k )={(1,1,1),(2,2,2),(3,3,3),(4,4,3),(5,4,5)}.

[0115] In order to evaluate the similarity of two trajectories, the dynamic time warping (DTW) algorithm is used and calculated according to the following formula:

[0116]

[0117] The trajectory contains K = 5 time points. The Euclidean distance of each time point is calculated in turn, and the similarity measure of the two trajectories is obtained by summing them up:

[0118] The distance at the first time point is:

[0119]

[0120] The distance at the second time point is:

[0121]

[0122] The distance at the third time point is:

[0123]

[0124] The distance at the fourth time point is:

[0125]

[0126] The distance at the fifth time point is:

[0127]

[0128] The total distance is:

[0129]

[0130] The calculation results show that the similarity distance between the two trajectories is 4.41. Combined with the similarity threshold (usually in the range of 0≤D≤10), it shows that the dynamic behaviors of the two trajectories have moderate similarity.

[0131] Based on similarity analysis, these fragmented potential trajectories are clustered with other fragmented potential data in the atrium, and the high-density distribution of fragmented potential areas are marked by the clustering results. Assume that the clustering results in Mr. Zhang's atrium show that the area with dense distribution of fragmented potentials is mainly concentrated near the right pulmonary vein and has a high migration frequency, which is a highly sensitive area for postoperative recurrence. By combining the trajectory similarity and spectral characteristics of this area (such as the average frequency f k =4Hz, corresponding to the time window T k =1s high active potential), which is further annotated and tracked to quantify the potential contribution of this area to postoperative recurrence and generate a high-risk area sensitivity map.

[0132] Coefficient value range

[0133] Coordinate units: x, y, z range is 0 ≤ x, y, z ≤ 20 cm;

[0134] The total number of time points K: usually 3≤K≤10;

[0135] Similarity threshold D: 0≤D≤10 is a reasonable range;

[0136] Time window T k :Take 0.5s≤T k ≤2s.

[0137] Through example analysis and computational verification, this method can not only effectively extract the dynamic characteristics of fragmentation potential, but also identify high-risk areas through trajectory similarity and cluster analysis.

[0138] In order to further quantify the recurrence sensitivity of the region, the migration intensity distribution model is used to model the sensitive region, generate a recurrence sensitivity distribution map, and output a regional characteristic report. The following is a detailed implementation process and data substitution calculation instructions.

[0139] The fragmentation potential trajectory data recorded by Mr. Zhang during the operation are as follows: In the time period t1 = 0.2s to t2 = 1.0s, there are N = 4 trajectory points in this area, which are Target point expressed the desire to quantify its recurrence sensitivity.

[0140] Migration intensity distribution model calculation:

[0141] According to the formula:

[0142]

[0143] Assume that the influence range adjustment parameter σ = 2, and the time difference between trajectory points is Δt = 0.2s. Calculate the Euclidean distance between the target point and the trajectory point point by point and substitute it into the formula.

[0144] right

[0145]

[0146] right

[0147]

[0148] right

[0149]

[0150] right

[0151]

[0152] Add up the contributions of each point:

[0153]

[0154] Assume that the time integration interval is t1 = 0.2s to t2 = 1.0s, and the time length is t2t1 = 0.8s.

[0155] M(x,y,z)=8.03×0.8=6.42

[0156] The calculation results show that the migration intensity distribution value of the target point (3,3,3) is M(3,3,3)=6.42, indicating that this point is located in a higher risk area. By modeling the sensitivity distribution of the entire three-dimensional space of the atrium, a recurrence sensitivity heat map is generated, showing that the vicinity of the right pulmonary vein is an area with a high intensity of fragmentation potential migration. Combined with the characteristic report, Mr. Zhang's recurrence risk is concentrated in this area, and it is recommended to strengthen the monitoring of electrophysiological activity in this area after surgery.

[0157] Coefficient value range

[0158] σ: influence range adjustment parameter, usually 1≤σ≤5;

[0159] Coordinate unit: x, y, z range 0 ≤ x, y, z ≤ 20 cm;

[0160] Time difference: Δt = t i+1 t i , take 0.1s≤Δt≤1.0s.

[0161] This example illustrates the application and feasibility of the migration intensity distribution model in the assessment of postoperative recurrence risk, and verifies the logic and practicality of the calculation results.

[0162] Embodiment 2:

[0163] Taking patient Mr. Zhang as an example, in order to evaluate the impact of fibrosis on atrial function after surgery, the three-dimensional reconstruction and erosion modeling method of atrial fibrosis was used to analyze the cardiac MRI image data at different time points (3 months, 6 months, and 1 year) after surgery, and record the dynamic expansion characteristics of the fibrosis area, including volume expansion rate, boundary migration trajectory, and interaction characteristics with low voltage areas and healthy tissues, as follows:

[0164] Step 1: 3D reconstruction and feature extraction of fibrosis areas

[0165] The cardiac MRI image of Mr. Zhang 3 months after surgery showed some fibrosis in the right atrium. After 3D reconstruction, the spatial distribution points of the fibrosis area were defined as Contains several 3D coordinate points, for example (x, y, z) distributed in the range of (2, 3, 4) to (5, 6, 7). Through reconstruction, the volume of the fibrosis area is calculated to be 15 cm 3 , the fiberization density function ρ(x,y,z) represents the degree of fiberization per unit volume. Assuming that the fiberization density is uniformly distributed in (x,y,z), ρ(x,y,z)=0.8.

[0166] Step 2: Calculation of fibrosis volume expansion rate

[0167] In the imaging data 6 months after surgery, the same fibrosis area The volume expands to 18cm 3 Assume that the time interval Δt = t2-t1 = 3 months, that is, 3 months × 30 days × 24 hours = 2160 hours. According to the formula:

[0168]

[0169] Substitute the data:

[0170]

[0171] The results showed that the volume expansion rate of the fibrotic area was 0.00111cm 3 / hour.

[0172] Step 3: Tracking the migration trajectory of fibrosis boundaries

[0173] Record the coordinate changes of several key points on the boundary of the fibrotic area. Assume that between 3 months and 6 months after surgery, the boundary point (3,3,5) migrates to (4,4,6). By calculating the Euclidean distance between two points:

[0174]

[0175] Migration speed of boundary points:

[0176]

[0177] The results showed that the migration speed of the boundary point was 0.0008 cm / hour, indicating that the fibrosis boundary was expanding slowly.

[0178] Step 4: Interaction analysis with neighboring areas

[0179] The extension of the fibrotic area caused it to overlap with the adjacent low-voltage area, and the volume of the overlapping area was calculated to be 5 cm 3 , accounting for 20% of the original low-voltage area. At the same time, the fibrosis extension path showed that it eroded into the healthy atrial tissue to a depth of 1.5 cm, further aggravating the signal abnormality of the low-voltage area.

[0180] The calculation results show that Mr. Zhang's fibrosis area continued to expand after surgery, with a volume expansion rate of 0.00111cm 3 / hour, and the boundary migration speed was 0.0008cm / hour. The expansion of fibrosis significantly eroded the low-voltage area and healthy tissue, with the erosion depth reaching 1.5cm. Combining these analyses, a report on the erosion characteristics of the fibrosis area was generated, marking high-risk areas to provide a scientific basis for postoperative review and intervention.

[0181] Coefficient value range:

[0182] 1. Density ρ(x,y,z): generally 0.5≤ρ≤1.0, reflecting the degree of fibrosis.

[0183] 2. Volume Normal range

[0184] 3. Time interval Δt: 1month≤Δt≤12months.

[0185] 4. Migration distance: typical range

[0186] Through example analysis and computational verification, this method has high feasibility and clinical applicability in quantifying fibrosis extension and erosion behavior, providing quantitative support for postoperative recurrence prediction and risk management.

[0187] Based on the previous analysis of the volume expansion and boundary migration of the fibrosis area, this stage focuses on analyzing the erosion behavior of the fibrosis area on the low-voltage area, including the quantification of boundary overlap features, the classification of erosion behavior, and the construction of a dynamic coupling model to evaluate the degree of damage of fibrosis expansion to the electrophysiological function of the low-voltage area.

[0188] Three-dimensional overlap between fibrosis and low-voltage areas in cardiac MRI images taken 6 months after surgery The extracted points cover the coordinates (x, y, z) from (2, 3, 4) to (5, 6, 6). The overlapping area A = 8 cm 2 , indicating that the fibrosis area partially invades the low voltage area in space. The erosion depth function κ(x,y,z) was calculated by image segmentation technology, and the average erosion depth was 2.5mm, indicating that the fibrosis area caused obvious damage to the boundary of the low voltage area.

[0189] According to the formula:

[0190]

[0191] The average value of the component function n(x, y, z) is 0.8, indicating that the corrosion behavior mainly extends to the interior of the low voltage area. The total interface area S = 12 cm 2 The mean value of the erosion depth function κ(x,y,z) is taken as 0.25cm. Substitute into the calculation:

[0192]

[0193] I r (t)=0.25·0.8·12=2.4cm 3 / hour

[0194] The results showed that the erosion rate of the fibrosis area to the low voltage area was 2.4 cm 3 / hour, which is a high erosion risk.

[0195] Based on the overlapping area and erosion characteristics, erosion behavior is classified into the following three cases:

[0196] P1 edge erosion: At 3 months after surgery, the fibrotic area was only close to the low-voltage area, the overlapping area was small, the erosion depth was less than 1 mm, and there was no significant functional damage.

[0197] P2 full coverage erosion: At 6 months after surgery, the fibrotic area completely covered part of the low-voltage area, and the erosion depth reached 2.5 mm, interfering with the signal conduction in the low-voltage area.

[0198] P3 Healthy tissue expansion: The image shows that the expansion path of the fibrotic area has invaded the healthy atrial tissue, resulting in a new low voltage area of ​​about 3cm 2 , forming new recurrence-sensitive areas.

[0199] A dynamic coupling model was constructed by combining the spatial expansion of the fibrosis area and the electrophysiological characteristics of the low-voltage area (such as voltage amplitude and signal fluctuation). Data analysis showed that in the erosion area, the voltage amplitude dropped from 0.8mV to 0.5mV, and the signal fluctuation amplitude increased by 30%, indicating that the expansion of fibrosis significantly weakened the signal conduction function of the low-voltage area. The dynamic coupling model was used to calculate the damage weight of the erosion area on the function of the low-voltage area, and the results showed that the contribution rate of this area to the risk of postoperative recurrence reached 45%.

[0200] Based on the calculation results, a characteristic report was generated, indicating that the erosion rate of Mr. Zhang's fibrosis area was 2.4cm 3 / hour, the erosion behavior is mainly full coverage erosion, accompanied by the expansion of healthy tissue, forming new low-voltage areas. The areas with higher risk of recurrence are concentrated near the top of the right atrium and the right pulmonary vein. It is recommended to strengthen electrophysiological monitoring of this area after surgery and combine imaging follow-up to evaluate the trend of fibrosis expansion.

[0201] Coefficient value range

[0202] 1. Erosion depth κ(x,y,z): Typical range is 0.1cm≤κ≤0.5cm.

[0203] 2. Normal component n(x,y,z): 0.5≤n≤1.0, indicating the degree of inward or outward erosion direction.

[0204] 3. Overlap area S: range is 5cm 2 ≤S≤20cm 2 .

[0205] Through case analysis, the effectiveness of this method in quantifying the erosion behavior of fibrotic areas and predicting the risk of recurrence was verified, which provides a scientific basis for precise postoperative intervention and has high clinical application value.

[0206] This example further quantifies the impact of fibrosis expansion on the reduction of low voltage area and the transformation of healthy tissue into low voltage area, records the dynamic changes of expansion rate, and marks high-risk areas based on erosion rate index. The following is a detailed analysis and example calculation process: 3 months after surgery, the initial area of ​​Mr. Zhang's low voltage area was Over time, fibrosis expanded and eroded the low voltage area, reducing it to approximately At the same time, the images showed that fibrosis eroded healthy tissue to form new low-voltage areas, with an additional area of ​​2cm 2 , indicating that healthy tissues are more susceptible to fibrosis extension.

[0207] The fibrosis expansion rate is defined and calculated by the erosion rate index E(t), which is used to reflect the changes in the expansion speed at different stages. The analysis period is T = 3 months, that is, t0 = 3 months, t0 + T = 6 months. The area change weight function γ(x, y, z) = 1 (uniform distribution assumption), substitute into the formula:

[0208]

[0209] The area of ​​fibrosis erosion changes over time:

[0210]

[0211] The reduction in area Substitute the data:

[0212]

[0213] Points calculation:

[0214]

[0215] Erosion Rate Index:

[0216]

[0217] The results showed that the erosion rate index of fibrosis expansion was E(t)=0.85, reflecting that the overall degree of damage to the low-voltage area caused by fibrosis expansion was high. Through the analysis of the dynamic changes in the expansion rate, Mr. Zhang's fibrosis expansion can be divided into a rapid expansion stage and a relatively stable stage. 3 to 4 months after surgery, the area reduction rate reached 20%, which was a rapid expansion stage; 4 to 6 months after surgery, the area reduction rate dropped to 10%, which was a relatively stable stage. Combined with the erosion depth (average 2.5 mm) and the area of ​​the newly added low-voltage area, the top of the right atrium and the vicinity of the right pulmonary vein were marked as high-risk areas for erosion.

[0218] The generated characteristic report pointed out that the postoperative fibrosis expansion of Mr. Zhang caused significant damage to the low voltage area, with an erosion rate index of 0.85. The top of the right atrium and the vicinity of the right pulmonary vein were high-risk areas, and the area of ​​the newly added low voltage area was 2cm 2 It is recommended to continuously monitor the fibrosis expansion trend in this area after surgery and optimize the intervention plan in combination with electrophysiological testing.

[0219] Coefficient value range

[0220] 1. Weight function γ(x,y,z): 0.5≤γ≤2, reflecting the unevenness of regional contribution.

[0221] 2. Initial area Range is

[0222] 3. Analysis period T: 1month≤T≤6months.

[0223] Through example calculation and characteristic analysis, the feasibility of this method in quantifying the impact of fibrosis expansion and predicting the risk of recurrence was verified, providing important quantitative basis and clinical guidance for postoperative management.

[0224] Embodiment 3:

[0225] This example uses Mr. Li, a patient without recurrence, as an example. The patient's atrial electrophysiological characteristics and dynamic behavior of fragmented potentials are analyzed through the intraoperative three-dimensional electroanatomical map and multimodal data of fragmented potential distribution. Combining the key data in the two images, it explains in detail how to predict the risk of postoperative recurrence and generate a low-risk assessment result through the method of the present invention.

[0226] Step 1: Reconstruction of 3D electroanatomical map and analysis of low voltage areas:

[0227] During the operation, the three-dimensional anatomical data of the patient's left atrium was collected, and a low-voltage area distribution map was generated based on electrophysiological mapping. The image shows that Mr. Li's low-voltage area is 10.9 cm 2 , accounting for 138cm of the total left atrium area 2The low voltage area is mainly distributed in the posterior atrial wall area, with a limited distribution range and stable voltage level. This small low voltage area and limited distribution suggest that Mr. Li has a low risk of recurrence after surgery.

[0228] Figure 4 :3D electroanatomical image and low voltage area measurement:

[0229] This figure shows the 3D electroanatomy of the left atrium. The red dotted line marks the range of the low voltage area, and the area of ​​the low voltage area (10.9 cm 2 ) and accounted for (7.9%). The low voltage area is mainly concentrated in the middle and lower part of the posterior wall of the atrium, and no large-scale diffusion is observed. The area and distribution of the low voltage area are important indicators for assessing the risk of recurrence. Mr. Li's low voltage area is relatively small and limited in distribution, which, combined with the low-risk characteristics, provides a strong basis for subsequent analysis.

[0230] Step 2: Analysis of spatial distribution and dynamic behavior of fragmentation potential

[0231] Through intraoperative mapping, the distribution and dynamic behavior of the patient's fragmented potentials were further analyzed. The images showed that the fragmented potentials were mainly concentrated in the anterior part of the right lower lung, with a short dynamic migration path, and no large-scale high-frequency active behavior. In addition, the voltage signal of the fragmented potential was relatively stable with a small fluctuation amplitude. This low-activity fragmented potential distribution further supports the conclusion that the patient has a low risk of postoperative recurrence.

[0232] Figure 5 :Dynamic distribution and migration path of fragmentation potential: The picture uses arrows to mark the concentrated distribution area of ​​fragmentation potential in the front of the right lower lung, and intuitively shows its dynamic behavior of the short path through the migration trajectory. The color map of the dynamic distribution shows that there is no significant high-frequency signal active area in this area. The dynamic characteristics of fragmentation potential are an important basis for assessing the risk of postoperative recurrence. Mr. Li's fragmentation potential has low activity and a short migration path, which further verifies his low risk of recurrence.

[0233] from Figure 4-5 The following conclusions can be drawn intuitively from this picture:

[0234] 1. Distribution characteristics of low voltage area: The low voltage area of ​​the left atrium of the patient is small (10.9cm 2 ), accounting for 7.9%, and the distribution is limited to the middle and lower part of the posterior wall. This small range and stable distribution of low-voltage areas indicate that the patient has a lower risk of postoperative recurrence.

[0235] 2. Dynamic behavior of fragmented potentials: The fragmented potentials are concentrated in the anterior part of the right lower lung, with a short migration path and low dynamic behavior activity, indicating that the patient's atrial electrophysiological characteristics are stable after surgery and do not have the typical characteristics of a high risk of recurrence.

[0236] Through the analysis of two pictures, this example shows how to use the intraoperative low voltage area distribution map and the dynamic distribution map of the fragmentation potential to assess the risk of postoperative recurrence. The area and distribution characteristics of the low voltage area and the dynamic behavior of the fragmentation potential all indicate that Mr. Li is a low-risk patient. This method based on multimodal data analysis provides accurate prediction of postoperative recurrence and provides a scientific basis for clinicians to develop personalized postoperative follow-up plans.

[0237] Step 3: Dynamic recurrence sensitivity assessment based on multimodal data

[0238] Based on the intraoperative data, the present invention further dynamically analyzes the distribution of low voltage areas, the dynamic behavior of fragmentation potentials, and their correlation with the risk of recurrence through the fusion of multimodal data. By calculating the dynamic migration rate and frequency fluctuation of fragmentation potentials, as well as the stability index of the low voltage area, a postoperative recurrence sensitivity model is constructed.

[0239] 3.1 Quantitative analysis of dynamic characteristics of fragmentation potential

[0240] The dynamic time warping (DTW) algorithm was used to analyze the trajectory characteristics of the fragmented potentials and quantify their dynamic activity. Mr. Li's fragmented potential active area was limited to the anterior part of the right lower lung, and the average length of the dynamic path was 2.5 mm, which was significantly lower than that of high-risk patients (the path was usually more than 10 mm). The frequency fluctuation of the fragmented potential was in the range of 0.4 Hz-0.6 Hz, and no high-frequency characteristics were observed (the frequency fluctuation of typical high-risk patients exceeded 1.0 Hz. The trajectory of the fragmented potential was short and the frequency fluctuation was low, indicating that the potential activity was weak, further supporting the characteristics of the patient's low risk of postoperative recurrence.

[0241] 3.2 Low voltage area stability analysis

[0242] Comparing the intraoperative and postoperative follow-up data, the area change rate of the low voltage zone is an important indicator of recurrence risk. The area of ​​Mr. Li's low voltage zone was only reduced by 0.3cm at the 6-month follow-up after surgery. 2 The area change rate was 2.8%, which is much lower than that of high-risk patients (usually the change rate exceeds 20%). In addition, the location and morphology of the low-voltage area remained stable during the follow-up period, without obvious migration or expansion behavior. The low area change rate and stable morphology of the low-voltage area indicate that the postoperative atrial electrophysiological activity was not significantly disturbed, further verifying that the patient has a low risk of recurrence after surgery.

[0243] Step 4: Generation of recurrence sensitivity heatmap

[0244] Based on the low voltage area distribution, the dynamic characteristics of the fragmented potential and its coupling analysis with the overall electrophysiological characteristics of the atrium, a recurrence sensitivity heat map was generated. The heat map intuitively presents the spatial distribution of recurrence risk through a color gradient (green indicates low risk and red indicates high risk).

[0245] Heatmap results:

[0246] 1. Risk assessment of low voltage area: The low voltage area is concentrated on the posterior wall, the area is stable, and no high-risk characteristics are observed. The heat map shows that this area is green, indicating a low risk of recurrence.

[0247] 2. Fragmentation potential activity assessment: The fragmentation potential activity in the anterior part of the right lower lung is low, the migration path is short, and it does not extend to healthy tissue. The heat map shows that this area is green to light green, indicating that the risk of recurrence in the potential active area is low.

[0248] The heat map integrates the dynamic analysis results of multimodal data into a visualization tool to help clinicians intuitively identify high-risk areas. For Mr. Li, the heat map shows that all atrial areas are green, with no significant high-risk features.

[0249] Step 5: Personalized recurrence probability curve and postoperative management plan:

[0250] Based on the dynamic analysis of the fragmentation potential and low voltage area, combined with the postoperative follow-up data, the postoperative recurrence probability curve of the patient was generated. The recurrence probability gradually decreased from 8% 3 months after surgery to 2% 1 year after surgery. The steady downward trend of the curve is consistent with the recurrence sensitivity thermoreactivity. Figure 1 This further verifies that the patient has a low risk of recurrence after surgery.

[0251] Personalized postoperative management recommendations:

[0252] 1. Follow-up period: It is recommended to conduct imaging review at 6 and 12 months after surgery to dynamically monitor changes in low-voltage areas and fibrosis areas. 2. Electrophysiological signal monitoring: During the period of 3 to 6 months after surgery, the monitoring frequency of fragmentation potentials and low-voltage areas should be appropriately increased to confirm their stability. 3. Recurrence risk intervention: Due to the extremely low risk of recurrence, no drug adjustment or secondary surgical intervention is required for the time being, only regular follow-up observation is required.

[0253] Through the analysis of two pictures and postoperative data, this embodiment comprehensively demonstrates the application process of the method for predicting recurrence after radiofrequency ablation of atrial fibrillation based on multimodal analysis of the present invention. Mr. Li's low voltage zone distribution range is small and stable, and the fragmentation potential migration path is short and low in activity. Combined with the dynamic analysis of erosion behavior and fibrosis expansion, a recurrence sensitivity heat map and a personalized recurrence probability curve are finally generated, successfully identifying him as a low-risk patient. This method significantly improves the accuracy of recurrence prediction through multimodal data fusion and dynamic modeling, and provides a scientific basis for postoperative management and follow-up optimization.

Claims

1. A method for predicting recurrence of atrial fibrillation after radiofrequency ablation based on multimodal analysis, characterized in that The following steps are involved: S1. Dynamic mapping and classification of intraoperative low voltage area characteristics: S1.

1. Use three-dimensional electroanatomical images to dynamically capture the morphology and spatial distribution of low-voltage areas; introduce regional complexity quantitative indicators to analyze the nonlinearity of low-voltage area boundaries; S1.

2. Classify the voltage intensity gradient within the low-voltage area and quantify the sensitivity to postoperative recurrence; establish a composite index model of low-voltage area proportion and fragmentation potential density; S1.

3. Based on spatial distribution and voltage characteristics, classify low voltage areas into high, medium, and low recurrence risks and output classification heat maps; S2. Spatial migration pattern and spectrum characteristics analysis of fragmentation potential: S2.

1. Analyze the migration path of intraoperative fragmentation potential in three-dimensional space, extract the migration speed, direction and frequency changes; use trajectory clustering algorithm to find the concentration trend of fragmentation potential in high recurrence sensitive areas; S2.

2. Extract the spectral characteristics of the fragmentation potential signal based on wavelet transform and identify the abnormal distribution of high-frequency fragmentation potential; design a frequency and space coupling model to integrate the temporal dynamics and spatial distribution of the fragmentation potential; S2.

3. Use feature selection algorithms including LASSO to screen key features of fractionated potentials associated with postoperative recurrence risk; S3. 3D reconstruction and erosion modeling of atrial fibrosis areas: S3.

1. Reconstruct the three-dimensional geometry of the fibrotic region based on postoperative cardiac MRI or CT data; extract the surface area, volume and distribution density of the fibrotic region; S3.

2. Analyze the erosion trajectory of the fibrotic area on the adjacent low-voltage area and healthy tissue; introduce the erosion rate index to quantify the degree of damage of fibrosis to the electrophysiological characteristics of the atrial S3.

3. Construct a recurrence sensitivity distribution map of the fibrosis area and overlay it with the coupling area of ​​the low voltage area for analysis; S4. Prediction of postoperative recurrence risk based on time series modeling: S4.

1. Extract time series features including voltage dynamic changes and fibrosis extension from intraoperative electrophysiological data and postoperative imaging follow-up data; use long short-term memory network (LSTM) to capture the dynamic change pattern of time series; S4.

2. Combine historical data to predict the time window for postoperative recurrence; output dynamic recurrence probability curve to provide time-sensitive risk assessment; S5. Recurrence prediction model based on multimodal feature fusion: S5.

1. Design a multimodal feature alignment network to unify intraoperative electrophysiological features, postoperative imaging features, and clinical features; use an adaptive weight allocation algorithm to optimize the feature alignment effect; S5.

2. A deep graph neural network (GNN) is used to model the interaction between low voltage areas, fragmentation potentials, and fibrosis areas. The model outputs personalized recurrence probability and generates a sensitivity map of high-risk areas.

2. The method for predicting recurrence of atrial fibrillation after radiofrequency ablation based on multimodal analysis according to claim 1, characterized in that The spatial migration pattern and spectrum characteristic analysis method of the fragmentation potential includes: The instantaneous velocity is calculated through the spatial displacement between adjacent time points to identify the high dynamic area; the directional characteristic data set of potential movement is constructed based on the change of three-dimensional spatial vector; the high active area and the stable area are distinguished by the change of activation frequency within a fixed time window, and the migration speed is calculated by the following formula: in: v i is the instantaneous speed of the i-th fragmentation potential, describing the i to i+1 The movement rate between i ,y i ,z i is the fragmentation potential at time t i The three-dimensional space coordinates at the time; i+1 ,y i+1 ,z i+1 is the fragmentation potential at time t i+1 The three-dimensional space coordinates at time (t i+1 t i ) is the time difference between two time points; by calculating the spatial distance between two adjacent coordinate points and combining the time difference, the movement speed of the potential is quantified.

3. The method for predicting recurrence of atrial fibrillation after radiofrequency ablation based on multimodal analysis according to claim 2, characterized in that The spatial migration pattern and spectrum characteristic analysis method of the fragmentation potential includes: The path of each fragmentation potential is abstracted into a three-dimensional trajectory curve, including the time dimension. The dynamic time warping (DTW) algorithm is used to calculate the similarity between trajectories, highlighting the comprehensive characteristics of migration speed, direction, and frequency. Based on the clustering results, the areas with high density distribution of fragmentation potentials are marked to quantify the potential contribution to postoperative recurrence. The trajectory similarity measurement formula is: in: For the trajectory and The dynamic similarity between i,k ,y i,k ,z i,k For the trajectory The three-dimensional space coordinates at the kth time point; x j,k ,y j,k ,z j,k For the trajectory The three-dimensional spatial coordinates at the kth time point; K is the total number of time points included in the trajectory.

4. The method for predicting recurrence of atrial fibrillation after radiofrequency ablation based on multimodal analysis according to claim 3, characterized in that The spatial migration pattern and spectrum characteristic analysis method of the fragmentation potential includes: The sensitive areas in the clustering results were modeled to analyze the concentration trend and directional distribution of the fragmentation potential. The recurrence sensitivity distribution map was constructed based on the density and migration intensity of the fragmentation potential in the sensitive area. The characteristic report of the recurrence sensitivity area was generated based on the heat map results, and the migration intensity distribution model was calculated using the following formula: in: M(x,y,z) is the migration intensity distribution of the target point (x,y,z); is the three-dimensional coordinate of the target point; is the three-dimensional coordinate of the trajectory point; σ is the adjustment parameter of the influence range, which controls the diffusion degree of the heat map distribution; t1, t2 are the time intervals of analysis; t i+1 t i is the time difference between adjacent time points.

5. The method for predicting recurrence of atrial fibrillation after radiofrequency ablation based on multimodal analysis according to claim 1, characterized in that The three-dimensional reconstruction and erosion modeling method of the atrial fibrosis region includes: Based on cardiac MRI or CT images, the three-dimensional morphology of the fibrosis area is reconstructed and the distribution characteristics of the boundary points are extracted. The image data at multiple time points including 3 months, 6 months and 1 year after surgery are analyzed to record the expansion rate of the volume of the fibrosis area and the boundary migration trajectory. By tracking the motion trajectory of the fibrosis boundary points, the contact points and expansion paths with the adjacent low-voltage areas and healthy tissues are identified.

6. The method for predicting recurrence of atrial fibrillation after radiofrequency ablation based on multimodal analysis according to claim 5, characterized in that The three-dimensional reconstruction and erosion modeling method of the atrial fibrosis region includes: The overlap between the fibrosis area and the low voltage area in three-dimensional space was compared to quantify the erosion behavior of erosion depth, area and density changes. A dynamic coupling model was established by combining the electrophysiological characteristics of the fibrosis area and the low voltage area, including voltage amplitude and signal fluctuation, to evaluate the degree of damage to the function of the low voltage area caused by fibrosis extension. The erosion rate of the fibrosis area and the low voltage area was described by the following formula: in: I r (t) is the erosion rate at time t, indicating the degree of damage of the fiberized area to the boundary of the low voltage area; is the interface between the fibrosis area and the low voltage area, which is the overlapping boundary of the two areas in three-dimensional space; κ(x, y, z) is the erosion depth function, which represents the erosion depth at the interface point (x, y, z); n(x, y, z) is the normal component function, which represents the distribution characteristics of the erosion direction on the interface, and is used to distinguish inward or outward erosion behavior; dS is a small area element on the interface, which is used to integrate and calculate the erosion rate of the entire interface.

7. The method for predicting recurrence of atrial fibrillation after radiofrequency ablation based on multimodal analysis according to claim 6, characterized in that The erosion behavior classification includes: P1, edge erosion: the boundary of the fibrosis area gradually approaches the low voltage area, but does not completely cover it; P2, full coverage erosion: the fibrosis area completely covers the low voltage area and further expands; P3. Healthy tissue expansion: Fibrosis erodes into healthy tissue areas, resulting in secondary generation of low voltage areas.

8. The method for predicting recurrence of atrial fibrillation after radiofrequency ablation based on multimodal analysis according to claim 6, characterized in that The three-dimensional reconstruction and erosion modeling method of the atrial fibrosis region includes: Quantify the effect of fibrosis expansion on the reduction of low voltage areas or the transformation of healthy tissue into low voltage areas; record the dynamic changes in the rate of fibrosis expansion, and identify the rapid stage and relatively stable stage of expansion; based on the expansion speed and erosion depth, mark the areas with high erosion risk as high-risk areas; the erosion rate index is defined as: in: E(t) is the erosion rate index at time t, reflecting the overall degree of damage to the low voltage area caused by fiber expansion; T is the length of the analysis period; is the area of ​​the fiberization erosion region at time t, indicating the degree of influence of the low voltage area; γ(x, y, z) is the area change weight function, describing the contribution of a certain point in the fiberization region to the overall erosion; is the area of ​​the initial low-voltage zone, which is used as a normalized benchmark to represent the area of ​​the low-voltage zone before surgery; dA is a small area element, which is used to integrally calculate the impact of the fibrosis area on the entire low-voltage zone.

Images