A PICC catheter tip precise navigation and fixation method and system based on real-time electrocardiogram positioning

Abstract

The application discloses a PICC catheter tip precise navigation and fixation method and system based on real-time electrocardiogram positioning, realizes high-precision positioning and safe fixation of the catheter tip through multi-modal data fusion and deep learning technology, and comprises the following steps: collecting intracavity electrocardio signals (ECG) through a built-in electrode of the catheter, combining with optical fiber sensor to obtain blood vessel wall contact pressure and temperature data; adopting an improved wavelet threshold algorithm to dynamically suppress motion artifacts and enhance P wave characteristics; inputting the preprocessed ECG signals into a time-space feature fusion Transformer model, using a time attention layer to analyze P wave time sequence changes, a space attention layer to integrate multi-lead space distribution characteristics, and combining with preoperative blood vessel dissection data to optimize positioning accuracy; fusing the ECG positioning results and the optical fiber sensing data through Kalman filtering to generate three-dimensional position information of the catheter tip, and realizing real-time navigation in an augmented reality interface; when the P wave amplitude reaches the CAJ threshold, the contact pressure is safe and there is no abnormal temperature, a shape memory alloy (SMA) fixation ring is triggered to contract, and precise fixation is realized. The application solves the problems that the traditional method depends on X-rays and has poor anti-interference performance, has the advantages of no radiation, high real-time performance and self-adaptive blood vessel deformation, and is suitable for clinical P catheter placement.

Description

Technical Field

[0001] This invention relates to the field of medical device technology, and in particular to a method and system for precise navigation and fixation of PICC catheter tip based on real-time electrocardiogram positioning, especially a real-time positioning and fixation system for PICC catheter tip that combines Transformer neural network model and fiber optic sensing technology. Background Technology

[0002] Peripherally inserted central venous catheters (PICCs) are widely used for long-term intravenous infusion, chemotherapy, and nutritional support. Traditional PICC placement relies on X-ray fluoroscopy or postoperative chest X-rays to confirm the tip position, which presents the following problems: Insufficient real-time capability: The catheter position cannot be adjusted in real time during the procedure, easily leading to misplacement (such as misplacement in the internal jugular vein or subclavian vein). Radiation exposure: Frequent X-ray verification increases the radiation risk for both medical staff and patients. Individual differences: Variations in vascular anatomy (such as tortuosity or abnormal branching) increase the difficulty of localization.

[0003] Current navigation technologies based on intracardiac electrocardiography (ECG) determine whether the catheter tip is close to the superior vena cava-atrial junction (SVC / CAJ) by observing changes in P-wave morphology. However, these technologies have the following limitations: Signal interference: Motion artifacts and electromyographic noise during catheter advancement reduce the accuracy of P-wave feature extraction. Insufficient multimodal data fusion: Existing methods do not fully integrate ECG with vascular biomechanical feedback (such as pressure and deformation), resulting in poor adaptability to dynamic vascular deformations (such as respiration and heartbeat). Weak generalization of the localization model: Traditional algorithms (such as thresholding methods) are difficult to adapt to individual differences in vascular anatomy and are easily affected by variations in P-wave amplitude.

[0004] In recent years, multi-sensor fusion and deep learning have provided new approaches for precision navigation: Transformer models: perform well in temporal signal processing, but have not yet been used for joint analysis of ECG spatial-temporal features. Fiber optic sensing technology: can monitor catheter-vessel contact mechanics (pressure, temperature) in real time, but lacks a framework for co-optimization with ECG. Augmented reality navigation: requires combining high-precision positioning and dynamic path planning to reduce the operator's workload.

[0005] To address the aforementioned issues, a method for precise navigation and fixation of the PICC catheter tip is needed. This method overcomes the shortcomings of existing technologies through the following innovations, achieving radiation-free, high-precision, and real-time interactive PICC catheter navigation, thus providing a safer and more efficient operating solution for clinical practice. Summary of the Invention

[0006] In view of this, it is necessary to provide a method and system for precise navigation and fixation of PICC catheter tip based on real-time ECG positioning, which mainly solves the problems of insufficient positioning accuracy of PICC catheter tip, lack of real-time navigation and passive fixation method in the existing technology.

[0007] In a first aspect, embodiments of this application provide a method for precise navigation and fixation of a PICC catheter tip based on real-time electrocardiogram positioning, the method comprising:

[0008] S1: Signal acquisition and preprocessing: The intracavitary electrocardiogram (ECG) signal is acquired in real time through the PICC catheter-embedded electrode, and the temperature sensing data of the first fiber optic component and the pressure sensing data of the second fiber optic component in the simulated blood vessel assembly are acquired simultaneously. The ECG signal is then filtered and P-wave feature enhancement is performed.

[0009] S2: Tip localization based on Transformer model: The preprocessed ECG signal is input into the Transformer model that fuses spatiotemporal features. The temporal changes in P-wave amplitude and morphology are analyzed through the temporal attention layer, and the spatial distribution features of multi-lead ECG are integrated through the spatial attention layer. Based on the P-wave features output by the model and the preset SVC / CAJ position mapping relationship, the relative distance and azimuth offset between the catheter tip and the target position are calculated.

[0010] S3: Multimodal real-time navigation: Combining the positioning results of the Transformer model with the sensing data of the first and second fiber optic components, data fusion is performed through Kalman filtering to generate three-dimensional position information of the catheter tip; the catheter advancement path and tip position are displayed in real time through an augmented reality navigation interface, and dynamic adjustment suggestions are provided;

[0011] S4: Intelligent fixation trigger: The catheter fixation procedure is initiated when the following conditions are met simultaneously: the P-wave amplitude output by the Transformer model reaches the CAJ position threshold; the blood vessel wall contact pressure detected by the second fiber optic component is within a safe range; and the first fiber optic component does not detect any abnormal temperature rise.

[0012] S5: Shape Memory Alloy SMA Fixation: Electrical stimulation is applied to the SMA fixation ring at the proximal end of the catheter to cause it to contract. The contraction force is dynamically adjusted according to the real-time pressure feedback of the second fiber optic assembly until the catheter tip is stable in the target position.

[0013] Optionally, in one implementation of the first aspect of the present invention, S2: Tip localization based on the Transformer model: The preprocessed ECG signal is input into a Transformer model that fuses spatiotemporal features. The temporal changes in P-wave amplitude and morphology are analyzed through a temporal attention layer, and the spatial distribution features of multi-lead ECGs are integrated through a spatial attention layer. Based on the P-wave features output by the model and the preset SVC / CAJ position mapping relationship, the relative distance and azimuth offset between the catheter tip and the target position are calculated, including:

[0014] S2.1: Preprocessing of multi-lead ECG signals, including noise reduction, R-peak identification, and heartbeat segmentation;

[0015] S2.2: The preprocessed ECG signal is input into the Transformer model for spatiotemporal feature fusion, where:

[0016] The time attention layer is used to analyze the temporal changes in P-wave amplitude and morphology, and to extract features in the time dimension;

[0017] The spatial attention layer is used to integrate the spatial distribution features of multi-lead ECG and extract features in the spatial dimension.

[0018] S2.3: Based on the P-wave characteristics output by the Transformer model and the preset SVC / CAJ position mapping relationship, calculate the relative distance and azimuth offset between the catheter tip and the target position;

[0019] S2.4: Based on the relative distance and azimuth offset, output the real-time positioning result of the catheter tip.

[0020] Optionally, in one implementation of the first aspect of the present invention, the preprocessing in S2.1 employs an improved wavelet threshold denoising algorithm, the specific formula of which is:

[0021]

[0022]

[0023] Where, x denoised (t) represents the time-domain expression of the denoised ECG signal at time t, where t represents the time variable. The wavelet coefficients after thresholding, w j,k ψ represents the wavelet decomposition coefficients of the original ECG signal. j,k (t) is the wavelet function of the time variable t, λ j To incorporate an adaptive threshold that combines motion artifacts and intravascular noise characteristics during catheter advancement, γ is the motion sensitivity coefficient, and the catheter jitter amplitude is estimated in real time using fiber optic pressure data. j Let J be the noise standard deviation of the j-th layer decomposition, J represent the total number of decomposition layers, N be the signal length / number of sampling points, and motion_artifact(t) represent the ECG signal noise component caused by the duct motion at time point t.

[0024] When based on fiber optic sensing data, the formula for motion_artifact(t) is:

[0025] motion_artifact(t)=||fiber_strain(t)||2·pressure_variance(t),

[0026] Wherein, fiber_strain(t) represents the catheter strain detected by the fiber optic deformation sensor, reflecting the amplitude of mechanical motion, and pressure_variance(t) represents the time-domain variance of the contact pressure on the blood vessel wall, reflecting the contact stability;

[0027] When a micro-inertial sensor is integrated into the catheter based on accelerometer signals, the formula for motion_artifact(t) is:

[0028] motion_artifact(t)=ξ·accel_RMS(t),

[0029] Here, accel_RMS(t) represents the root mean square value of the proximal catheter acceleration, and ξ represents the adjustment coefficient; the P-wave signal-to-noise ratio is enhanced to ensure that the ECG signal input to the Transformer model can accurately reflect the electrophysiological characteristics of the catheter tip and the CAJ at the junction of the superior vena cava (SVC) and atrium.

[0030] Optionally, in one implementation of the first aspect of the present invention, the time attention layer in S2.2 calculates the P-wave time series characteristics through a masked self-attention mechanism, specifically using the following formula:

[0031]

[0032] Among them, Q t =X t W q K t =X t W k V t =X t W v These are the query, key, and value matrix, respectively, X. t For input P-wave amplitude-morphology sequence, W q W k W v Let d be the weight matrix. k For the time-attention head dimension, it is selected based on the characteristics of the ECG signal, M t This is a diagonal mask matrix, adjusted according to the heartbeat period, used to mask information about future moments. The output temporal feature is the result of concatenating multiple attention heads, denoted as F. t =MultiHead(Q t ,K t V t ).

[0033] Optionally, in one implementation of the first aspect of the present invention, the spatial attention layer in S2.2 integrates spatial features through cross-guided attention, specifically as follows:

[0034]

[0035] Among them, X s This is a multi-lead ECG signal matrix, where the rows represent the number of leads and the columns represent the characteristic dimensions of a single lead. W a W b W c Let F be the spatial attention weight matrix, and let F be the output spatial feature. s =SpatialAttn(X s ), d s For the spatial attention head dimension, A θ δ represents the vascular orientation matrix, derived from preoperative CT / MRI, where δ is the anatomical weight, forcing the model to focus on lead features consistent with the main vascular orientation.

[0036] Optimize spatial attention by leveraging prior knowledge of vascular anatomy to reduce localization bias caused by individual vascular variations.

[0037] Optionally, in one implementation of the first aspect of the present invention, the spatiotemporal feature fusion described in 2.2 is achieved through feature concatenation and gating mechanisms, specifically as follows:

[0038] F fusion =α·F t +(1-α)·F s ,

[0039] Where, α=σ(W g [F t ;F s ]+b g ) represents the dynamic fusion weights, [F t ;F s ] represents the time feature F t Spatial characteristics F s splicing, W g b g Here are the weights and bias parameters of the gating mechanism, σ is the sigmoid activation function, and F... fusion The fused features are the final input to the Transformer decoder;

[0040] Optimize spatial attention by leveraging prior knowledge of vascular anatomy to reduce localization bias caused by individual vascular variations.

[0041] Optionally, in one implementation of the first aspect of the present invention, the preset SVC / CAJ position mapping relationship in S2.3 is constructed through a Gaussian process regression model, and the calculation formulas for the relative distance D of the catheter tip and the azimuth offset θ are as follows:

[0042]

[0043] k D ,k θ =λ1k RBF (x,x')+λ2k Matern (x,x')+λ3k dir (x,x'),

[0044] Among them, a composite kernel function was designed to capture the branching structure and curvature changes of blood vessels, including the RBF kernel function k. RBF (x,x'), used for global smoothness adaptation; Matern kernel function k matern (x,x'), used for local curvature adaptation; direction-sensitive kernel function k dir (x,x') is used for vascular branch constraints, and the formula is:

[0045]

[0046] Where, k D (·), k θ (·) is the Gaussian process kernel function based on the fitting of training samples, Θ vessel ε represents vascular anatomy parameters. D ε θ It is zero-mean Gaussian noise, and its variance is negatively correlated with the confidence level of fiber optic sensing, with variances of respectively. Dynamically adjust according to signal quality, and satisfy (x, y) are the coordinates of the catheter tip, (x target y target ), θ=arctan2(yy target ,xx target ) represents the target position coordinates of SVC / CAJ, λ1, λ2, and λ3 represent the trainable combined weights learned from surgical data, and A θ The direction weight matrix is ​​adjusted according to the main blood vessel direction SVC→CAJ path, l is the blood vessel length scale, and x and x' represent two data points in the input space.

[0047] The vessel diameter gradually narrows from SVC to CAJ. Dynamic kernel parameters are set, and the length scale l is set as a position-dependent function. Real-time fiber optic deformation data is introduced to correct the curvature. The formula is:

[0048] l(x)=l0·(1+μ·curvature(x)+η·fiber_strain(x)),

[0049] Where curvature(x) is the local curvature of the blood vessel, which is pre-calculated using preoperative CT / MRI data, μ is an adjustable hyperparameter that controls the curvature sensitivity, l(x) represents the dynamic length scale, l0 represents the initial length scale, fiber_strain(x) is the fiber strain feedback, and η is the deformation sensitivity coefficient.

[0050] The kernel function parameters are dynamically adjusted to adapt the localization model to the physical deformation of blood vessels during surgery, including vascular displacement caused by respiration or heartbeat.

[0051] Optionally, in one implementation of the first aspect of the present invention, wherein S3: multimodal real-time navigation: combining the positioning results of the Transformer model with the sensing data of the first and second fiber optic components, data fusion is performed through Kalman filtering to generate three-dimensional position information of the catheter tip; the catheter advancement path and tip position are displayed in real time through an augmented reality navigation interface, and dynamic adjustment suggestions are provided, including:

[0052] The positioning results output by the Transformer model are spatiotemporally aligned with the strain, temperature, and deformation data collected by the first and second fiber optic components.

[0053] The Kalman filter algorithm is used to fuse multi-source data, calculate the optimal three-dimensional position estimate of the catheter tip, and output the position confidence score.

[0054] The fused navigation data is input into the augmented reality navigation interface to render the three-dimensional advancement path and tip position of the catheter in real time. At the same time, combined with vascular anatomy data, a visual ruler is used to indicate the deviation between the current position and the target area.

[0055] Based on path planning algorithms and mechanical feedback data, dynamic duct adjustment suggestions are generated, including propulsion direction correction, speed adjustment threshold, and risk warning information.

[0056] Secondly, embodiments of this application provide a PICC catheter tip precision navigation and fixation system based on real-time electrocardiogram positioning, applied to the PICC catheter tip precision navigation and fixation method based on real-time electrocardiogram positioning as described in the first aspect, the system comprising:

[0057] The signal acquisition and preprocessing module is used to acquire intracavitary electrocardiogram (ECG) signals in real time through the PICC catheter-embedded electrodes, and simultaneously acquire temperature sensing data of the first optical fiber component and pressure sensing data of the second optical fiber component in the simulated blood vessel assembly, and perform filtering and P-wave feature enhancement processing on the ECG signals.

[0058] The tip localization module based on the Transformer model is used to input the preprocessed ECG signal into the Transformer model that fuses spatiotemporal features. It analyzes the temporal changes in P-wave amplitude and morphology through a temporal attention layer and integrates the spatial distribution features of multi-lead ECG through a spatial attention layer. Based on the P-wave features output by the model and the preset SVC / CAJ position mapping relationship, it calculates the relative distance and azimuth offset between the catheter tip and the target position.

[0059] The multimodal real-time navigation module is used to combine the positioning results of the Transformer model with the sensing data of the first and second fiber optic components, perform data fusion through Kalman filtering, and generate three-dimensional position information of the catheter tip; the module displays the catheter advancement path and tip position in real time through an augmented reality navigation interface and provides dynamic adjustment suggestions.

[0060] The intelligent fixation trigger module is used to initiate the catheter fixation procedure when the following conditions are met simultaneously: the P-wave amplitude output by the Transformer model reaches the CAJ position threshold; the blood vessel wall contact pressure detected by the second fiber optic component is within a safe range; and the first fiber optic component does not detect an abnormal temperature rise.

[0061] The shape memory alloy SMA fixation module is used to apply electrical stimulation to the SMA fixation ring at the proximal end of the catheter to cause it to contract. The contraction force is dynamically adjusted according to the real-time pressure feedback of the second fiber optic assembly until the catheter tip is stable in the target position.

[0062] Thirdly, embodiments of this application provide an electronic device, including:

[0063] processor;

[0064] Memory used to store processor-executable instructions;

[0065] The processor is configured to implement the precise navigation and fixation method for PICC catheter tip based on real-time ECG positioning as described in the first aspect when executing the instructions.

[0066] Fourthly, embodiments of this application provide a computer-readable storage medium storing a program that instructs a device to perform the PICC catheter tip precision navigation and fixation method based on real-time electrocardiogram positioning as described in the first aspect.

[0067] This invention discloses a method and system for precise navigation and fixation of PICC catheter tips based on real-time electrocardiogram (ECG) positioning. It achieves high-precision positioning and safe fixation of the catheter tip through multimodal data fusion and deep learning technology. The method includes: real-time acquisition of intracavitary electrocardiogram (ECG) signals via an internal electrode, combined with fiber optic sensors to obtain vascular wall contact pressure and temperature data; dynamic suppression of motion artifacts and enhancement of P-wave features using an improved wavelet thresholding algorithm; inputting the preprocessed ECG signal into a Transformer model that fuses spatiotemporal features, using a temporal attention layer to analyze the temporal changes of the P-wave, and a spatial attention layer to integrate multi-lead spatial distribution features, combined with preoperative vascular anatomy data to optimize positioning accuracy; fusing ECG positioning results and fiber optic sensor data through Kalman filtering to generate three-dimensional position information of the catheter tip, and providing real-time navigation on an augmented reality interface; triggering the contraction of a shape memory alloy (SMA) fixation ring when the P-wave amplitude reaches the CAJ threshold, the contact pressure is safe, and there is no abnormal temperature, achieving precise fixation. This invention solves the problems of traditional methods relying on X-rays and having poor anti-interference capabilities, and has the advantages of being radiation-free, highly real-time, and adaptable to vascular deformation, making it suitable for clinical P-wave catheter placement.

[0068] Beneficial effects:

[0069] (1) Improved positioning accuracy: By using an improved wavelet threshold denoising algorithm and a Transformer model that fuses spatiotemporal features, motion artifacts are effectively suppressed and P-wave feature extraction is enhanced, thus significantly improving the positioning accuracy of the catheter tip.

[0070] (2) Achieve radiation-free navigation: Real-time positioning is achieved entirely based on ECG signals and fiber optic sensing data, avoiding the radiation hazards of traditional X-ray positioning, making it safer and more environmentally friendly.

[0071] (3) Enhanced real-time performance: Multimodal data fusion technology is adopted, combined with Kalman filtering algorithm, to realize real-time tracking and dynamic adjustment of catheter position, which significantly improves surgical efficiency.

[0072] (4) Adapting to individual differences: By optimizing the spatial attention mechanism through prior knowledge of vascular anatomy and dynamically adjusting model parameters in combination with intraoperative fiber optic sensing data, the vascular variations of different patients can be effectively addressed.

[0073] (5) Enhanced safety: The intelligent fixation triggering mechanism takes into account multiple parameters such as P wave characteristics, contact pressure and temperature to ensure that the catheter tip is stably fixed in the target position and avoid the risk of vascular damage.

[0074] (6) Easy to operate: The augmented reality navigation interface intuitively displays the catheter position and advancement path, and provides real-time adjustment suggestions, which greatly reduces the difficulty of operation for operators.

[0075] (7) High system integration: The system modularly integrates functions such as signal acquisition, processing, navigation and fixation to form a complete intelligent surgical assistance system, which is convenient for clinical application and promotion. Attached Figure Description

[0076] Figure 1 This is a schematic diagram of a method for precise navigation and fixation of the PICC catheter tip based on real-time electrocardiogram positioning, provided in an embodiment of this application.

[0077] Figure 2 This is a diagram illustrating the architecture of a PICC catheter tip precision navigation and fixation system based on real-time electrocardiogram positioning, provided as an embodiment of this application.

[0078] Figure 3 A schematic diagram of an electronic device provided in an embodiment of this application. Detailed Implementation

[0079] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them.

[0080] It should be noted that, in the embodiments of this application, "at least one" refers to one or more, and "more than one" refers to two or more. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used in the specification of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of this application.

[0081] It should be noted that in the embodiments of this application, the terms "first," "second," etc., are used only for descriptive purposes and should not be construed as indicating or implying relative importance, nor as indicating or implying order. Features specified as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of the embodiments of this application, words such as "exemplary" or "for example" are used to indicate examples, illustrations, or explanations. Any embodiment or design scheme described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design schemes. Specifically, the use of words such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner.

[0082] Based on the embodiments described in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0083] Example 1

[0084] The present invention proposes a PICC catheter tip precision positioning system that integrates multimodal sensing, intelligent algorithms and dynamic fixation. Through five core steps, it realizes full-process intelligentization from signal acquisition to automatic fixation, which significantly improves the safety and accuracy of catheter insertion. Figure 1 This is a schematic flowchart illustrating a method for precise navigation and fixation of the PICC catheter tip based on real-time electrocardiogram positioning, provided in one embodiment of this application. Figure 1 As shown, a method for precise navigation and fixation of PICC catheter tip based on real-time electrocardiogram localization includes:

[0085] S1: Signal Acquisition and Preprocessing: The intracavitary electrocardiogram (ECG) signal is acquired in real time through the PICC catheter-embedded electrode. Temperature sensing data of the first fiber optic component and pressure sensing data of the second fiber optic component in the simulated blood vessel assembly are acquired simultaneously. The ECG signal is then filtered and P-wave feature enhancement is performed.

[0086] Specifically, the first fiber optic component monitors vascular temperature to screen for thermal damage or infection risk in the tissue surrounding the catheter; the second fiber optic component acquires the contact pressure with the vessel wall, reflecting the physical interaction between the catheter tip and the vessel wall. The preprocessing stage involves filtering and denoising the ECG signal (e.g., removing power frequency interference and electromyographic noise) and enhancing P-wave features (highlighting the amplitude and morphological details of the P wave through wavelet transform or adaptive filtering), providing key biomarkers with a high signal-to-noise ratio for subsequent localization. This enables coordinated monitoring of multiple parameters—ECG, temperature, and pressure—laying the foundation for accurate localization and safe fixation.

[0087] S2: Tip localization based on Transformer model: The preprocessed ECG signal is input into the Transformer model that fuses spatiotemporal features. The temporal attention layer analyzes the temporal changes in P-wave amplitude and morphology, and the spatial attention layer integrates the spatial distribution features of multi-lead ECG. Based on the P-wave features output by the model and the preset SVC / CAJ position mapping relationship, the relative distance and azimuth offset between the catheter tip and the target position are calculated.

[0088] A spatiotemporal feature fusion Transformer model is used to process preprocessed ECG signals, overcoming the limitations of traditional P-wave localization methods that rely on single leads or static features: A temporal attention layer focuses on the dynamic changes in P-wave amplitude as the catheter advances (e.g., characteristic attenuation in the right atrium → superior vena cava (SVC) → interatrial junction (CAJ) region), capturing temporal correlations; a spatial attention layer integrates the spatial distribution differences of multi-lead ECGs, enhancing adaptability to the diversity of P-wave morphology in different anatomical locations (e.g., complementary signal features between right ventricular leads and coronary sinus leads). The P-wave features output by the model are used to calculate the relative distance (millimeter-level) and azimuth offset (three-dimensional coordinate deviation) between the tip and the target location (e.g., CAJ) through a pre-defined SVC / CAJ position mapping relationship (a P-wave parameter-anatomical position regression model constructed based on clinical big data).

[0089] Transformer's self-attention mechanism enables adaptive weight allocation of long-range temporal dependencies and spatial features, resulting in a significant improvement in positioning accuracy compared to traditional algorithms.

[0090] The structure diagram of the Transformer model includes, in sequence: input embedding layer, position encoding layer, temporal attention layer, spatial attention layer, spatiotemporal feature fusion module, Transformer encoder-decoder stack, and position mapping output layer.

[0091] The Input Embedding Layer converts the raw ECG signal (continuous time series) into a high-dimensional vector representation for easier model processing. Its input is the preprocessed ECG signal (dimension: number of leads × time step, e.g., 12 leads × 1000 sampling points). The embedding matrix embeds each sampling point into a fixed-dimensional vector (e.g., 256 dimensions) through a linear transformation.

[0092] The Position Encoding Layer adds temporal location information to help the model recognize the temporal order of ECG signals (such as the specific time point of P wave occurrence). Position encoding vectors are generated using sine and cosine functions and superimposed on the embedding vector. For the spatial dimension, prior knowledge of vascular anatomy (such as the vascular orientation matrix from preoperative CT / MRI) is introduced to add spatial encoding to lead positions (e.g., the relative positional weights of leads I, II, and III). Position encoding is crucial for Transformer's processing of sequence data, enabling the model to understand the order of elements.

[0093] The Temporal Attention Layer is used to analyze the temporal changes in P-wave amplitude and morphology, capturing the patterns of feature changes over time (such as the dynamics of the P-wave's rising and falling phases). Masked Self-Attention is used to calculate attention weights at each time step, emphasizing important moments (such as the P-wave peak). Multi-Head Attention uses multiple attention heads in parallel (e.g., 8 heads) to output a concatenated temporal feature vector (e.g., capturing periodicity and morphological shifts). Similar to temporal modeling in action localization, the Transformer handles long-range dependencies through self-attention.

[0094] The Spatial Attention Layer integrates the spatial distribution features of multi-lead ECGs, reducing the impact of individual vascular variations (e.g., optimizing attention to leads aligned with the main vascular direction). Cross-Lead Attention calculates attention weights between leads based on vascular anatomy priors (e.g., direction-sensitive kernel functions). The output spatial feature vector enhances the model's perception of lead spatial relationships (e.g., the symmetry features between leads aVR and aVL).

[0095] The Spatio-Temporal Fusion Module dynamically fuses temporal and spatial features to generate a comprehensive representation input to the Transformer decoder. Inputs: Outputs from the temporal attention layer (temporal feature vectors) and the spatial attention layer (spatial feature vectors). Fusion strategy: Concatenation: Concatenates the two types of feature vectors along their respective dimensions. Gating mechanism: Dynamically adjusts the fusion ratio using learnable weights (e.g., sigmoid gating units), prioritizing more important features. Output: The fused feature vector, input to subsequent Transformer layers.

[0096] The Transformer encoder-decoder stack is used to further process the fused features, generating a high-level P-wave feature representation. Multiple encoder layers: Each layer contains a self-attention sublayer and a feed-forward neural network, with residual connections and layer normalization added. Decoder layer (optional): Used if the model requires sequence generation, but typically simplified in this localization task (encoder only). Output: P-wave feature vector (e.g., 512-dimensional), containing temporal and spatial information.

[0097] The Position Mapping Output Layer calculates the relative distance and azimuth offset of the catheter tip based on the P-wave characteristics output by the model and the preset SVC / CAJ mapping relationship. Mapping Model: Based on Gaussian Process Regression (GPR), using composite kernel functions (such as RBF kernel, Matern kernel) to capture vascular structures (branches, curvature). Calculation Output: Relative Distance: Scalar value (e.g., distance from SVC / CAJ in millimeters). Azimuth Offset: Vector (e.g., three-dimensional offset angle). Dynamic Adjustment: Incorporating real-time fiber optic deformation data correction parameters (e.g., vascular deformation caused by respiration).

[0098] Specifically, S2: Tip localization based on the Transformer model: The preprocessed ECG signal is input into a Transformer model that fuses spatiotemporal features. The temporal changes in P-wave amplitude and morphology are analyzed through a temporal attention layer, and the spatial distribution features of multi-lead ECGs are integrated through a spatial attention layer. Based on the P-wave features output by the model and the preset SVC / CAJ position mapping relationship, the relative distance and azimuth offset between the catheter tip and the target position are calculated, including:

[0099] S2.1: Preprocessing of multi-lead ECG signals, including denoising, R-peak identification, and heartbeat segmentation. Multi-lead ECG signals provide comprehensive information for cardiovascular disease diagnosis by synchronously acquiring electrical activity from different parts of the heart (e.g., a 12-lead ECG can reflect the spatial distribution characteristics of myocardial ischemia and arrhythmias). However, ECG signals are susceptible to interference from power frequency (50Hz / 60Hz), electromyographic noise (EMG), baseline drift (caused by respiratory or postural changes), and electrode contact noise, leading to waveform distortion (e.g., misjudgment of ST segment shift and blurring of QRS complexes). Traditional wavelet threshold denoising performs well in single-lead signals, but multi-lead signals exhibit noise correlation (e.g., synchronous power frequency interference) and lead specificity (e.g., differences in signal-to-noise ratio between different leads), requiring targeted improvements to enhance overall denoising performance.

[0100] In ECG signal processing, preprocessing is a crucial step in ensuring the accuracy of subsequent analysis. Denoising can be achieved through digital filtering techniques, such as using bandpass filters to remove high-frequency noise and low-frequency drift. R-peak identification typically relies on the characteristics of the QRS complex, while heartbeat segmentation requires dividing the continuous ECG signal into individual heartbeat cycles. These steps can be accomplished using custom algorithms or deep learning-based methods.

[0101] Specifically, the preprocessing in S2.1 employs an improved wavelet threshold denoising algorithm, with the following formula:

[0102]

[0103] Where, x denoised (t) represents the time-domain expression of the denoised ECG signal at time t, where t represents the time variable. The wavelet coefficients after thresholding, w j,k ψ represents the wavelet decomposition coefficients of the original ECG signal. j,k (t) is the wavelet function of the time variable t, λ j To incorporate an adaptive threshold that combines motion artifacts and intravascular noise characteristics during catheter advancement, γ is the motion sensitivity coefficient, and the catheter jitter amplitude is estimated in real time using fiber optic pressure data. j Let J be the noise standard deviation of the j-th layer decomposition, J represent the total number of decomposition layers, N be the signal length / number of sampling points, and motion_artifact(t) represent the ECG signal noise component caused by the duct motion at time point t.

[0104] When based on fiber optic sensing data, the formula for motion_artifact(t) is:

[0105] motion_artifact(t)=||fiber_strain(t)||2·pressure_variance(t),

[0106] Wherein, fiber_strain(t) represents the catheter strain detected by the fiber optic deformation sensor, reflecting the amplitude of mechanical motion, and pressure_variance(t) represents the time-domain variance of the contact pressure on the blood vessel wall, reflecting the contact stability;

[0107] When a micro-inertial sensor is integrated into the catheter based on accelerometer signals, the formula for motion_artifact(t) is:

[0108] motion_artifact(t)=ξ·accel_RMS(t),

[0109] Here, accel_RMS(t) represents the root mean square value of the proximal catheter acceleration, and ξ represents the adjustment coefficient; the P-wave signal-to-noise ratio is enhanced to ensure that the ECG signal input to the Transformer model can accurately reflect the electrophysiological characteristics of the catheter tip and the CAJ at the junction of the superior vena cava (SVC) and atrium.

[0110] To enhance the P-wave signal-to-noise ratio and ensure that the ECG signal input to the Transformer model accurately reflects the electrophysiological characteristics of the catheter tip and the superior vena cava (SVC) / atrial junction (CAJ), an improved wavelet thresholding denoising algorithm is employed. This algorithm combines motion artifacts during catheter advancement with adaptive threshold settings based on intravascular noise characteristics to effectively remove interference from motion noise and intravascular noise, thereby enhancing the P-wave signal-to-noise ratio. This ensures that the ECG signal input to the Transformer model accurately reflects the electrophysiological characteristics of the catheter tip and the SVC / CAJ, providing a reliable data foundation for subsequent accurate localization.

[0111] S2.2: The preprocessed ECG signal is input into a Transformer model that fuses spatiotemporal features. The temporal attention layer analyzes the temporal variations in P-wave amplitude and morphology, extracting features in the time dimension. The spatial attention layer integrates the spatial distribution features of multi-lead ECG, extracting features in the spatial dimension.

[0112] The temporal attention layer is used to analyze the temporal changes in P-wave amplitude and morphology, extracting features along the time dimension. Transformer models capture long-term dependencies in sequences through self-attention mechanisms, which is particularly important when processing ECG signals. For example, the ST-ECGFormer model effectively captures spatiotemporal features in ECG signals through token embedding, spatiotemporal and positional embedding, and a Transformer encoder.

[0113] The temporal attention layer calculates the P-wave temporal features through a masked self-attention mechanism, analyzes the changes in P-wave amplitude and morphology over time, and captures the differences in the importance of P-wave features at different times. The output temporal features are the result of splicing multiple attention heads, enhancing the model's ability to perceive dynamic changes in the P-wave and helping to accurately identify the temporal changes in P-wave features during the movement of the catheter tip within the blood vessel.

[0114] Specifically, in S2.2, the time attention layer calculates the P-wave temporal characteristics through a masked self-attention mechanism, using the following formula:

[0115]

[0116] Among them, Q t =X t W q K t =X t W k V t =X t W v These are the query, key, and value matrix, respectively, X. t For input P-wave amplitude-morphology sequence, W q Wk W v Let d be the weight matrix. k For the time-attention head dimension, it is selected based on the characteristics of the ECG signal, M t This is a diagonal mask matrix, adjusted according to the heartbeat period, used to mask information about future moments. The output temporal feature is the result of concatenating multiple attention heads, denoted as F. t =MultiHead(Q t ,K t V t ).

[0117] The temporal attention layer calculates the P-wave temporal features through a masked self-attention mechanism, analyzes the changes in P-wave amplitude and morphology over time, and captures the differences in the importance of P-wave features at different times. The output temporal features are the result of splicing multiple attention heads, enhancing the model's ability to perceive dynamic changes in the P-wave and helping to accurately identify the temporal changes in P-wave features during the movement of the catheter tip within the blood vessel.

[0118] Specifically, the spatial attention layer in S2.2 integrates spatial features through cross-lead attention, as shown in the following formula:

[0119]

[0120] Among them, X s This is a multi-lead ECG signal matrix, where the rows represent the number of leads and the columns represent the characteristic dimensions of a single lead. W a W b W c Let F be the spatial attention weight matrix, and let F be the output spatial feature. s =SpatialAttn(X s ), d s For the spatial attention head dimension, A θ The vascular orientation matrix is ​​derived from preoperative CT / MRI, and δ represents the anatomical weight. This forces the model to focus on lead features consistent with the main vascular orientation. Spatial attention is optimized through prior knowledge of vascular anatomy to reduce localization bias caused by individual vascular variations.

[0121] The spatial attention layer integrates spatial features through cross-lead attention, utilizing prior knowledge of vascular anatomy (such as the vascular orientation matrix and anatomical weights from preoperative CT / MRI) to optimize the focus on lead features consistent with the main vascular orientation. This reduces localization bias caused by individual vascular variations, improves the model's adaptability to differences in vascular structure among different patients, and enriches P-wave feature information from a spatial dimension.

[0122] The spatial attention layer is used to integrate the spatial distribution features of multi-lead ECGs and extract features in the spatial dimension. This method is similar to the spatiotemporal visual language models (RS-STVLMs) in remote sensing, which process multimodal data through a multi-head attention mechanism and a feedforward layer. In ECG signal processing, spatial attention can enhance information interaction between different leads, thereby improving positioning accuracy.

[0123] Specifically, the spatiotemporal feature fusion described in section 2.2 is achieved through feature concatenation and gating mechanisms, with the specific formula as follows:

[0124] F fusion =α·F t +(1-α)·F s ,

[0125] Where, α=σ(W g [F t ;F s ]+b g ) represents the dynamic fusion weights, [F t ;F s ] represents the time feature F t Spatial characteristics F s splicing, W g b g Here are the weights and bias parameters of the gating mechanism, σ is the sigmoid activation function, and F... fusion The fused features are the final input to the Transformer decoder;

[0126] Optimize spatial attention by leveraging prior knowledge of vascular anatomy to reduce localization bias caused by individual vascular variations.

[0127] Spatiotemporal feature fusion is achieved through feature concatenation and gating mechanisms, dynamically adjusting the fusion weights of temporal and spatial features to generate more comprehensive and representative fused features input to the Transformer decoder. Furthermore, prior knowledge of vascular anatomy is incorporated to optimize the fusion process, ensuring the model can comprehensively utilize the temporal variations and spatial distribution information of the P wave, thereby improving the accuracy and robustness of localization features.

[0128] S2.3: Based on the P-wave characteristics output by the Transformer model and the preset SVC / CAJ position mapping relationship, calculate the relative distance and azimuth offset between the catheter tip and the target position.

[0129] The P-wave characteristics output by the Transformer model can be mapped to preset SVC / CAJ locations. For example, intracardiac electrocardiography (IC-ECG) determines the catheter tip position by analyzing changes in the P-wave. When the catheter tip enters the SVC, the P-wave gradually increases; when the catheter tip reaches the CAJ, the positive P-wave increases significantly; and when the catheter tip passes the CAJ and enters the right atrium, a negative P-wave may appear. These characteristics allow for the calculation of the relative distance and azimuth offset between the catheter tip and the target location.

[0130] Specifically, the preset SVC / CAJ position mapping relationship in S2.3 is constructed using a Gaussian process regression model, and the calculation formulas for the relative distance D of the catheter tip and the azimuth offset θ are as follows:

[0131]

[0132] k D ,k θ =λ1k RBF (x,x')+λ2k Matern (x,x')+λ3k dir (x,x'),

[0133] Among them, a composite kernel function was designed to capture the branching structure and curvature changes of blood vessels, including the RBF kernel function k. RBF (x,x'), used for global smoothness adaptation; Matern kernel function k matern (x,x'), used for local curvature adaptation; direction-sensitive kernel function k dir (x,x') is used for vascular branch constraints, and the formula is:

[0134]

[0135] Where, k D (·), k θ (·) is the Gaussian process kernel function based on the fitting of training samples, Θ vessel ε represents vascular anatomy parameters. D ε θ It is zero-mean Gaussian noise, and its variance is negatively correlated with the confidence level of fiber optic sensing, with variances of respectively. Dynamically adjust according to signal quality, and satisfy (x, y) are the coordinates of the catheter tip, (x target y target ), θ=arctan2(yy target ,xx target ) represents the target position coordinates of SVC / CAJ, λ1, λ2, and λ3 represent the trainable combined weights learned from surgical data, and A θThe direction weight matrix is ​​adjusted according to the main blood vessel direction SVC→CAJ path, l is the blood vessel length scale, and x and x' represent two data points in the input space.

[0136] The vessel diameter gradually narrows from SVC to CAJ. Dynamic kernel parameters are set, and the length scale l is set as a position-dependent function. Real-time fiber optic deformation data is introduced to correct the curvature. The formula is:

[0137] l(x)=l0·(1+μ·curvature(x)+η·fiber_strain(x)),

[0138] Where curvature(x) is the local curvature of the blood vessel, which is pre-calculated using preoperative CT / MRI data, μ is an adjustable hyperparameter that controls the curvature sensitivity, l(x) represents the dynamic length scale, l0 represents the initial length scale, fiber_strain(x) is the fiber strain feedback, and η is the deformation sensitivity coefficient.

[0139] The kernel function parameters are dynamically adjusted to adapt the localization model to the physical deformation of blood vessels during surgery, including vascular displacement caused by respiration or heartbeat.

[0140] Based on a Gaussian process regression model, a pre-defined SVC / CAJ position mapping relationship is constructed. Composite kernel functions (RBF kernel, Matern kernel, and orientation-sensitive kernel) are designed to capture the branching structure, curvature changes, and directional constraints of blood vessels. Kernel function parameters are dynamically adjusted (e.g., position-dependent length scales, and the introduction of real-time fiber optic deformation data to correct curvature) to adapt the positioning model to the physical deformation of blood vessels caused by respiration, heartbeat, etc., during surgery, accurately calculating the relative distance and azimuth offset between the catheter tip and the target position.

[0141] Specifically, the dynamic adjustment method for the noise covariance of Kalman filtering can be: adjusting the confidence level of the fiber pressure / temperature data. Related to ECG signal quality: Among them, SNR ECG Here, ECG signal-to-noise ratio is represented, pressure_variance is the pressure fluctuation, and δ is the pressure fluctuation weight. This improves the robustness of data fusion and avoids navigation errors caused by the failure of a single sensor.

[0142] S2.4: Based on the relative distance and azimuth offset, output the real-time positioning result of the catheter tip. Based on the calculated relative distance and azimuth offset, the real-time positioning result of the catheter tip can be output. This step is similar to the position-time aware Transformer model in remote sensing change detection, where a change map is generated through feature fusion and a classifier head. In catheter positioning, the real-time feedback mechanism ensures precise control of the catheter tip, thereby improving the safety and effectiveness of catheter placement.

[0143] The S2.1 to S2.4 process covers the entire process from ECG signal preprocessing to real-time catheter tip positioning, combining traditional signal processing technology and advanced Transformer model to achieve high-precision catheter positioning.

[0144] Specifically, intelligent fixed trigger conditions can also perform multi-threshold collaborative optimization, formulating the fixed trigger conditions into a joint probability model: In this model, α1, α2, and α3 are weights, normalized weights for each parameter obtained through training with surgical data. These weights balance the contributions of electrophysiological, mechanical, and temperature parameters. For example, a higher α1 indicates greater reliance on P-wave positioning accuracy; a higher α2 prioritizes pressure safety; and α3, with a negative weight, suppresses the influence of abnormal temperatures. ΔT represents the rate of abnormal temperature change (e.g., frictional heating) detected by the first fiber optic assembly, used to prevent tissue thermal damage caused by catheter friction. Quantifying these parameters reduces the risk of misfixation, ensuring the catheter tip remains stable in the CAJ without damaging the vessel wall. wave This represents the real-time P-wave amplitude characteristic value output by the Transformer model, reflecting the electrophysiological proximity between the catheter tip and the CAJ (atrial junction). threshold This represents the preset P-wave amplitude threshold, calibrated using clinical data, and serves as a benchmark value for determining whether the catheter tip has reached the CAJ target position. current This indicates the real-time blood vessel wall contact pressure value detected by the second fiber optic component. (P) safe This indicates the upper limit of safe pressure that the blood vessel wall can withstand (to avoid perforation or damage). (Similar to P) current Mechanical safety when common constraints are fixed. σ(·) is the Sigmoid activation function. The comprehensive score is mapped to a fixed trigger probability of [0,1], with the threshold typically set to 0.5.

[0145] This model achieves the following optimizations by quantifying the synergistic effects of multiple parameters: Precise triggering: SMA fixation is initiated only when electrophysiological, mechanical, and temperature conditions are simultaneously met (P(fix)≥0.5), avoiding misoperation. Dynamic adaptation: Weighting coefficients can be adjusted according to individual patient differences (such as vascular fragility), improving safety.

[0146] Improvements in motion artifact adaptive denoising enhance the accuracy of P-wave feature extraction, reducing positioning failures caused by catheter jitter. Enhanced spatial attention based on vascular orientation constraints adapts to individual vascular anatomy variations, reducing the frequency of intraoperative X-ray verification. A dynamic kernel function based on fiber optic feedback responds to vascular deformation in real time, improving navigation stability. Multimodal noise covariance adjustment balances the reliability of ECG and fiber optic data, avoiding the risk of single sensor failure. Combined probabilistic triggering fixation integrates electrophysiological, mechanical, and temperature parameters, reducing perforation or displacement complications. These improvements not only mathematically optimize algorithm performance but also closely serve the two core objectives of precise navigation and safe fixation, meeting the clinical needs of PICC catheter placement.

[0147] S3: Multimodal real-time navigation: Combining the positioning results of the Transformer model with the sensing data of the first and second fiber optic components, data fusion is performed through Kalman filtering to generate three-dimensional position information of the catheter tip; the catheter advancement path and tip position are displayed in real time through an augmented reality navigation interface, and dynamic adjustment suggestions are given.

[0148] Integrating Transformer positioning results with fiber optic sensor data, dynamic fusion of multi-source information is achieved through Kalman filtering: ECG positioning provides anatomical location benchmarks, while fiber optic temperature / pressure data provides physical status feedback (e.g., abnormal pressure indicates a risk of catheter apposition to the vessel wall, requiring adjustment of the advancement direction); the fused data generates three-dimensional position coordinates, which are displayed in real-time through an augmented reality (AR) navigation interface: a three-dimensional reconstruction of the catheter advancement path (overlaid on the patient's CT / MRI images); dynamic deviation between the current tip position and the target area; and dynamic adjustment suggestions based on pressure / temperature thresholds (e.g., "fine-tune 2° to the left to reduce vessel wall pressure"). Clinical value: Transforming abstract physiological signals into intuitive operational guidance reduces operator reliance on experience and shortens operation time.

[0149] S3: Multimodal Real-time Navigation: Combining the positioning results of the Transformer model with the sensing data of the first and second fiber optic components, data fusion is performed using Kalman filtering to generate three-dimensional position information of the catheter tip; the catheter advancement path and tip position are displayed in real time through an augmented reality navigation interface, and dynamic adjustment suggestions are provided, including:

[0150] The positioning results output by the Transformer model are spatiotemporally aligned with the strain, temperature, and deformation data collected by the first and second fiber optic components.

[0151] The Kalman filter algorithm is used to fuse multi-source data, calculate the optimal three-dimensional position estimate of the catheter tip, and output the position confidence score.

[0152] The fused navigation data is input into the augmented reality navigation interface to render the three-dimensional advancement path and tip position of the catheter in real time. At the same time, combined with vascular anatomy data, a visual ruler is used to indicate the deviation between the current position and the target area.

[0153] Based on path planning algorithms and mechanical feedback data, dynamic duct adjustment suggestions are generated, including propulsion direction correction, speed adjustment threshold, and risk warning information.

[0154] S4: Intelligent fixation trigger: The catheter fixation procedure is initiated when the following conditions are met simultaneously: the P-wave amplitude output by the Transformer model reaches the CAJ position threshold; the blood vessel wall contact pressure detected by the second fiber optic component is within a safe range; and the first fiber optic component does not detect any abnormal temperature rise.

[0155] Specifically, three trigger conditions are set to ensure the accuracy and safety of fixation timing, avoiding premature or incorrect fixation: ECG condition: The P wave amplitude output by the Transformer model reaches the CAJ position threshold (e.g., a decrease of 40% ± 5% compared to the amplitude in the right atrium, with the specific value dynamically calibrated according to the patient's body shape); Pressure condition: The vascular wall contact pressure detected by the second fiber optic component is within a safe range (5-15 mmHg, avoiding high pressure leading to vascular damage or low pressure indicating tip drift); Temperature condition: No abnormal temperature rise is detected by the first fiber optic component (e.g., an increase of <2℃ from the baseline temperature, ruling out the risk of catheter-related thrombosis or infection).

[0156] By using AND logic, we ensure that all three conditions are met simultaneously, maximizing the effectiveness and security of fixed operations.

[0157] S5: Shape Memory Alloy SMA Fixation: Electrical stimulation is applied to the SMA fixation ring at the proximal end of the catheter to cause it to contract. The contraction force is dynamically adjusted according to the real-time pressure feedback of the second fiber optic assembly until the catheter tip is stable in the target position.

[0158] Specifically, a SMA fixation ring (integrated into the proximal end of the catheter) is used for mechanical fixation, replacing traditional suture or adhesive fixation. Electrical stimulation drive: A preset current (0.5-2A) is applied to the SMA ring, causing it to contract based on shape memory effect, generating radial support force to fix the catheter. Pressure closed-loop feedback: The pressure on the vessel wall after fixation is monitored in real time through a second fiber optic component, dynamically adjusting the SMA contraction force (e.g., reducing the current when the pressure exceeds 15 mmHg, and increasing the current when it is less than 5 mmHg). Stability endpoint judgment: When the pressure fluctuation is <2 mmHg and lasts for 3 seconds, the tip is considered stable at the target position, and adjustment is stopped. Material advantages: SMA has high biocompatibility and low elastic modulus, which can reduce long-term stimulation to the vessel wall; dynamic pressure feedback avoids the risk of vascular stenosis or perforation caused by over-fixation.

[0159] The core advantages of the solution include: 1. Multimodal fusion: Combining ECG physiological signals with fiber optic physical sensing, providing dual assurance of positioning accuracy and safety; 2. Intelligent algorithm-driven: The Transformer model breaks through the spatiotemporal limitations of traditional methods and adapts to individual anatomical differences; 3. Real-time visual navigation: The AR interface reduces operational complexity and enhances the surgeon's confidence; 4. Closed-loop dynamic fixation: SMA + pressure feedback achieves adaptive anchoring, avoiding the uncertainty of manual fixation.

[0160] Example 2

[0161] like Figure 2 As shown, this application provides an architecture diagram of a PICC catheter tip precision navigation and fixation system based on real-time ECG positioning, which is applied to the PICC catheter tip precision navigation and fixation system based on real-time ECG positioning as described in Embodiment 1. It includes a signal acquisition and preprocessing module 11, a tip positioning module 12 based on the Transformer model, a multimodal real-time navigation module 13, an intelligent fixation trigger module 14, and a shape memory alloy SMA fixation module 15.

[0162] The signal acquisition and preprocessing module 11 is used to acquire intracavitary electrocardiogram (ECG) signals in real time through the PICC catheter-embedded electrodes, simultaneously acquire temperature sensing data of the first optical fiber component and pressure sensing data of the second optical fiber component in the simulated blood vessel assembly, and perform filtering and P-wave feature enhancement processing on the ECG signals.

[0163] The tip localization module 12 based on the Transformer model is used to input the preprocessed ECG signal into the Transformer model that fuses spatiotemporal features. It analyzes the temporal changes of P wave amplitude and morphology through the temporal attention layer and integrates the spatial distribution features of multi-lead ECG through the spatial attention layer. Based on the P wave features output by the model and the preset SVC / CAJ position mapping relationship, it calculates the relative distance and azimuth offset between the catheter tip and the target position.

[0164] The multimodal real-time navigation module 13 is used to combine the positioning results of the Transformer model with the sensing data of the first and second fiber optic components, perform data fusion through Kalman filtering, and generate three-dimensional position information of the catheter tip; the augmented reality navigation interface displays the catheter advancement path and tip position in real time and provides dynamic adjustment suggestions.

[0165] The intelligent fixation trigger module 14 is used to start the catheter fixation procedure when the following conditions are met simultaneously: the P-wave amplitude output by the Transformer model reaches the CAJ position threshold; the blood vessel wall contact pressure detected by the second fiber optic component is within a safe range; and the first fiber optic component does not detect an abnormal temperature rise.

[0166] The shape memory alloy SMA fixation module 15 is used to apply electrical stimulation to the SMA fixation ring at the proximal end of the catheter to cause it to contract. The contraction force is dynamically adjusted according to the real-time pressure feedback of the second fiber optic assembly until the catheter tip is stable in the target position.

[0167] Figure 3 This is an electronic device provided in one embodiment of this application. For example... Figure 3 As shown, the electronic device includes at least the following components: processor 101 and memory 100, communication interface 103, and bus 102.

[0168] In this embodiment of the application, memory 100 is used to store executable instructions of processor 101, which, when configured to execute instructions, implements the method as described in the first aspect.

[0169] In embodiments of this application, a computer-readable storage medium includes instructions that instruct a device to perform the method as described in the first aspect. For example, the instructions instruct the device to perform... Figure 1 The method is shown in the process steps.

[0170] In one embodiment of this application, the program operating in the electronic device may be a program that controls a central processing unit (CPU) or similar device to achieve the functions of the above-described embodiments of the present invention (a program that enables the computer to function). Information processed by these devices is then temporarily stored in random access memory (RAM) during processing, and subsequently stored in various ROMs such as read-only memory (FlashROM) and hard disk drives (HDDs), and read, corrected, and written by the CPU as needed.

[0171] It should be noted that a portion of the electronic device described above can also be implemented using a computer. In this case, the program for implementing the control function can be recorded on a computer-readable recording medium, and the program recorded on the recording medium can be read into the computer and executed.

[0172] It should be noted that the term "computer" as used here refers to a computer built into an electronic device, employing hardware including an operating system and peripheral devices. Furthermore, "computer-readable recording media" refers to removable media such as floppy disks, magneto-optical disks, ROMs, and CD-ROMs, as well as storage devices such as hard drives built into a computer.

[0173] Furthermore, a "computer-readable recording medium" can include: a medium that dynamically stores a program for a short period of time, such as a communication line used when transmitting a program via a network such as the Internet or a communication line such as a telephone line; or a medium that stores a program for a fixed period of time, such as volatile memory inside a computer that serves as a server or client in this case. In addition, the aforementioned program can be a program used to implement the above-mentioned functions, or it can be a program that can implement the above-mentioned functions by combining with programs already recorded in the computer.

[0174] Furthermore, the electronic device in the above embodiments can also be implemented as an assembly (device group) composed of multiple devices. Each device constituting the device group can possess some or all of the functions or functional blocks of the electronic device in the above embodiments. As a device group, it is sufficient to have all the functions or functional blocks of the electronic device.

[0175] Those skilled in the art should recognize that the above embodiments are only used to illustrate this application and are not intended to limit this application. Any appropriate changes and variations made to the above embodiments within the essential spirit and scope of this application fall within the scope of protection claimed in this application.

Claims

1. A precise navigation and fixation system for PICC catheter tips based on real-time electrocardiogram positioning, characterized in that, The system includes: The signal acquisition and preprocessing module is used to acquire intracavitary electrocardiogram (ECG) signals in real time through the PICC catheter-embedded electrodes, and simultaneously acquire temperature sensing data of the first optical fiber component and pressure sensing data of the second optical fiber component in the simulated blood vessel assembly, and perform filtering and P-wave feature enhancement processing on the ECG signals. The tip localization module based on the Transformer model is used to input preprocessed ECG signals into a Transformer model that fuses spatiotemporal features. It analyzes the temporal variations of P-wave amplitude and morphology through a temporal attention layer and integrates the spatial distribution characteristics of multi-lead ECGs through a spatial attention layer. Based on the P-wave characteristics output by the model and the preset SVC / CAJ position mapping relationship, it calculates the relative distance and azimuth offset between the catheter tip and the target location. This includes: Preprocessing of multi-lead ECG signals, including noise reduction, R-peak identification, and heartbeat segmentation; The preprocessed ECG signal is input into the Transformer model for spatiotemporal feature fusion, where: The time attention layer is used to analyze the temporal changes in P-wave amplitude and morphology, and to extract features in the time dimension; The spatial attention layer is used to integrate the spatial distribution features of multi-lead ECG and extract features in the spatial dimension. Based on the P-wave characteristics output by the Transformer model and the preset SVC / CAJ position mapping relationship, the relative distance and azimuth offset between the catheter tip and the target position are calculated. Based on the relative distance and azimuth offset, the real-time positioning result of the catheter tip is output; The preprocessing employs an improved wavelet threshold denoising algorithm, with the specific formula as follows: ， ， ， in, Represented as the denoised ECG signal in time The time-domain expression at that point, Represents a time variable. These are the wavelet coefficients after thresholding. These are the wavelet decomposition coefficients of the original ECG signal. For time variables wavelet function, To incorporate an adaptive threshold that combines motion artifacts during catheter advancement with intravascular noise characteristics, The motion sensitivity coefficient is used to estimate the catheter vibration amplitude in real time using fiber optic pressure data. For the first The noise standard deviation of layer decomposition, Indicates the total number of levels in the decomposition. Signal length / number of sampling points Indicates at a point in time At that time, the noise component of the ECG signal caused by catheter movement; When based on fiber optic sensing data The formula is: ， in, This represents the strain in the conduit detected by the fiber optic deformation sensor, reflecting the amplitude of mechanical motion. The time-domain variance of the pressure at which the vessel wall contacts the vessel reflects the stability of the contact. When a micro-inertial sensor is integrated into the catheter based on accelerometer signals, The formula is: ， in, This is expressed as the root mean square value of the acceleration at the proximal end of the catheter. Indicates the adjustment factor; Enhance the P-wave signal-to-noise ratio to ensure that the ECG signal input to the Transformer model can accurately reflect the electrophysiological characteristics of the catheter tip and the CAJ at the junction of the superior vena cava (SVC) and atrium. The multimodal real-time navigation module is used to combine the positioning results of the Transformer model with the sensing data of the first and second fiber optic components, perform data fusion through Kalman filtering, and generate three-dimensional position information of the catheter tip; the module displays the catheter advancement path and tip position in real time through an augmented reality navigation interface and provides dynamic adjustment suggestions. The intelligent fixation trigger module is used to initiate the catheter fixation procedure when the following conditions are met simultaneously: the P-wave amplitude output by the Transformer model reaches the CAJ position threshold; the blood vessel wall contact pressure detected by the second fiber optic component is within a safe range; and the first fiber optic component does not detect an abnormal temperature rise. The shape memory alloy SMA fixation module is used to apply electrical stimulation to the SMA fixation ring at the proximal end of the catheter to cause it to contract. The contraction force is dynamically adjusted according to the real-time pressure feedback of the second fiber optic assembly until the catheter tip is stable in the target position.

2. The PICC catheter tip precision navigation and fixation system based on real-time electrocardiogram positioning according to claim 1, characterized in that, The time attention layer is used to analyze the temporal changes in P-wave amplitude and morphology, extracting features in the time dimension. The specific formula is as follows: ， in, , , These are query, key, and value matrices, respectively. To input the P-wave amplitude-morphology sequence, This is the weight matrix. The time-attention head dimension is selected based on the characteristics of the ECG signal. This is a diagonal mask matrix, adjusted according to the heartbeat period, used to mask future time information. The output temporal feature is the result of concatenating multiple attention heads, represented as follows: .

3. The PICC catheter tip precision navigation and fixation system based on real-time electrocardiogram positioning according to claim 1, characterized in that, The spatial attention layer is used to integrate the spatial distribution features of multi-lead ECG and extract features in the spatial dimension. The specific formula is as follows: ， in, This is a multi-lead ECG signal matrix, where the rows represent the number of leads and the columns represent the characteristic dimensions of a single lead. The spatial attention weight matrix is ​​used to output spatial features. , For spatial attention head dimension, This is a vascular orientation matrix, derived from preoperative CT / MRI. As an anatomical weight, the model is forced to focus on lead features that are consistent with the main direction of the blood vessel; Optimize spatial attention by leveraging prior knowledge of vascular anatomy to reduce localization bias caused by individual vascular variations.

4. The PICC catheter tip precision navigation and fixation system based on real-time electrocardiogram positioning according to claim 1, characterized in that, The spatiotemporal feature fusion is achieved through feature concatenation and gating mechanisms, with the specific formula as follows: ， in, Indicates dynamic fusion weights. Indicating time characteristics Spatial features splicing, The weights and bias parameters of the gating mechanism, It is the sigmoid activation function. The fused features are the final input to the Transformer decoder; Optimize spatial attention by leveraging prior knowledge of vascular anatomy to reduce localization bias caused by individual vascular variations.

5. A PICC catheter tip precision navigation and fixation system based on real-time electrocardiogram positioning according to claim 1, characterized in that, The preset SVC / CAJ position mapping relationship is constructed using a Gaussian process regression model, and the relative distance between the catheter tips is... and azimuth offset The calculation formula is: ， ， Among them, composite kernel functions were designed to capture the branching structure and curvature changes of blood vessels, including the RBF kernel function. Used for global smoothness adaptation; Matern kernel function Used for local curvature adaptation; direction-sensitive kernel function Used for vascular branch constraint, the formula is: ， ， ， in, The kernel function is a Gaussian process fitted based on the training samples. Indicates vascular anatomy parameters, It is zero-mean Gaussian noise, and its variance is negatively correlated with the confidence level of fiber optic sensing, with variances of respectively. Dynamically adjust according to signal quality, and meet the following requirements. , The coordinates of the catheter tip are: , The coordinates of the SVC / CAJ target location. This represents the trainable combined weights learned from surgical data. The direction weight matrix is ​​used to adjust the anisotropy according to the main blood vessel direction SVC→CAJ path. This is a measure of blood vessel length. This represents two data points in the input space; The vessel diameter gradually narrows from the SVC to the CAJ. Dynamic kernel parameters are set to adjust the length scale. Set as a position-dependent function, and introduce real-time fiber deformation data to correct curvature, the formula is: ， in, The local curvature of the blood vessels is pre-calculated using preoperative CT / MRI data. To adjust the hyperparameters and control curvature sensitivity, Indicates a dynamic length scale. Indicates the initial length scale. For fiber optic strain feedback, The deformation sensitivity coefficient; The kernel function parameters are dynamically adjusted to adapt the localization model to the physical deformation of blood vessels during surgery, including vascular displacement caused by respiration or heartbeat.

6. The PICC catheter tip precision navigation and fixation system based on real-time electrocardiogram positioning according to claim 1, characterized in that, in, The multimodal real-time navigation module is used to combine the positioning results of the Transformer model with the sensing data of the first and second optical fiber components, and perform data fusion through Kalman filtering to generate three-dimensional position information of the catheter tip. The augmented reality navigation interface displays the duct advancement path and tip position in real time and provides dynamic adjustment suggestions, including: The positioning results output by the Transformer model are spatiotemporally aligned with the strain, temperature, and deformation data collected by the first and second fiber optic components. The Kalman filter algorithm is used to fuse multi-source data, calculate the optimal three-dimensional position estimate of the catheter tip, and output the position confidence score. The fused navigation data is input into the augmented reality navigation interface to render the three-dimensional advancement path and tip position of the catheter in real time. At the same time, combined with vascular anatomy data, a visual ruler is used to indicate the deviation between the current position and the target area. Based on path planning algorithms and mechanical feedback data, dynamic duct adjustment suggestions are generated, including propulsion direction correction, speed adjustment threshold, and risk warning information.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a program that instructs the device to be used in the PICC catheter tip precision navigation and fixation system based on real-time electrocardiogram positioning as described in any one of claims 1 to 6.

Images