Puncture system for cardiovascular disease treatment

Abstract

The present application relates to the technical field of intelligent monitoring of medical devices, and discloses a puncture system for cardiovascular disease treatment. The system comprises a data receiving and preprocessing module, a signal mapping and path tracking module, a behavior pattern induction and deviation marking module, a traceability analysis and knowledge matching module, and a diagnosis and control instruction generation module. It extracts effective signal components through multi-stage noise suppression, establishes a time-event mapping network for them to track the signal evolution path, and then induces typical behavior patterns and marks deviations. The system performs traceability analysis on the deviation data points, traces back to the source signal characteristics and related events, and then matches them with the preset working logic knowledge base, finally generating descriptive diagnostic fragments and a set of targeted control instructions. The present application realizes intelligent behavior understanding and self-explanatory abnormal diagnosis of the puncture process, can accurately locate the root cause of the deviation and generate effective control strategies, and improves the safety and controllability of the operation.

Description

Technical Field

[0001] This invention relates to the field of intelligent monitoring technology for medical devices, specifically a puncture system for the treatment of cardiovascular diseases. Background Technology

[0002] In interventional cardiovascular procedures, real-time monitoring of the puncture process is crucial for ensuring surgical safety. Current technologies generally employ real-time monitoring methods based on fixed thresholds. The core of this approach is to filter the physiological signals collected by sensors and then compare the instantaneous values ​​or simple waveform characteristics with preset static thresholds. Once the signal exceeds the threshold, the system triggers an alarm. This method can only respond to isolated, instantaneous states of the signal.

[0003] This technique, which relies on instantaneous diagnosis, has inherent flaws. Puncture is a continuous, dynamic process, and signal changes are closely linked to specific operative events in a temporal causal relationship. Current technology severs this connection, failing to understand the current abnormal signal within its complete evolutionary path. This results in the system only being able to indicate "where the abnormality is," but not "why the abnormality is," meaning it cannot distinguish between signal changes that may appear similar instantaneously due to different root causes such as operational differences, tissue characteristics, or instrument contact. Diagnosis remains superficial, lacking the ability to trace the root cause of the abnormality, leading to subsequent regulatory recommendations that lack specificity and rely heavily on generalization based on the operator's experience.

[0004] Current monitoring technologies need to be upgraded from passive threshold alarms to proactive systems with process understanding and intelligent diagnostic capabilities. This invention requires a technical solution capable of constructing signal temporal evolution paths, associating operational events with signal characteristics, and performing logical tracing and matching diagnostics of deviation root causes based on domain knowledge. This aims to achieve a leap from "status alarms" to "root cause diagnosis," thereby providing a direct basis for precise control. Summary of the Invention

[0005] The purpose of this invention is to provide a puncture system for the treatment of cardiovascular diseases, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides a puncture system for the treatment of cardiovascular diseases, the system comprising: The data receiving and preprocessing module is used to receive real-time monitoring data of the puncture process from external devices, and to perform multi-level noise suppression on the real-time monitoring data of the puncture process to extract effective signal components characterizing the system operating status. The signal mapping and path tracing module is used to establish a time and event dimension mapping network for the effective signal components, and to track and record the evolution path of the effective signal components within a continuous time window through the mapping network. The behavior pattern induction and deviation marking module is used to inductively identify recurring typical behavior patterns based on the evolution path, and to compare the effective signal components within the current time window with the typical behavior patterns to mark the set of data points that exceed a preset deviation threshold. The source analysis and knowledge matching module is used to perform source analysis on the data point set to trace the source signal characteristics and time-related events that generated the data point set, and to match the source signal characteristics and time-related events with a preset puncture system working logic knowledge base; The diagnostic and control instruction generation module is used to generate a descriptive diagnostic fragment on the operating status of the puncture system based on the matching results, and to generate a set of control instructions for potential deviations in the operating status based on the descriptive diagnostic fragment.

[0007] Preferably, performing source tracing analysis on the data point set includes the following steps: Analyze the temporal and amplitude distribution characteristics of the data point set; Retrieve records of abnormal events with similar distribution characteristics from the historical monitoring database; Extract the precursor signal sequence from the abnormal event record; Calculate the morphological similarity between the data point set and each precursor signal sequence; Based on the morphological similarity, the target precursor signal sequence with the strongest reference value is selected; Analyze the historical fault event chain corresponding to the target precursor signal sequence; Extract the core trigger signal combination from the aforementioned chain of historical failure events; Search for traces of the core trigger signal combination in the real-time monitoring data of the current puncture procedure; If traces of the occurrence of the core trigger signal combination are found, the current form of the core trigger signal combination and its precise occurrence time are defined as source signal characteristics. Device operation events or physiological parameter events that are strongly correlated in time with the current form of the core trigger signal are defined as time-related events.

[0008] Preferably, calculating the morphological similarity between the data point set and each precursor signal sequence includes the following steps: The data point set and each precursor signal sequence are normalized to eliminate differences in absolute dimensions. For the normalized set of data points, calculate its statistical distribution moments; For each normalized precursor signal sequence, calculate its corresponding statistical distribution moments; Construct a similarity evaluation framework that includes multiple distance metrics, including at least distribution moment distance, dynamic time warping distance, and waveform structure distance; Using the aforementioned similarity assessment framework, the comprehensive distance between the data point set and each precursor signal sequence is calculated one by one; The comprehensive distance is converted into a morphological similarity score; Establish an index correspondence between morphological similarity scores and precursor signal sequences.

[0009] Preferably, constructing a similarity evaluation framework that includes multiple distance metrics includes the following steps: The calculation rules for the distribution moment distance are defined, which are achieved by comparing the differences between the data point set and the precursor signal sequence in terms of mean, variance, skewness, and kurtosis. A calculation rule for dynamic time warping distance is defined, which is achieved by finding the optimal nonlinear alignment path between the time series of the data point set and the time series of the precursor signal sequence, and calculating the cumulative distance under the optimal nonlinear alignment path; The calculation rules for waveform structure distance are defined. These rules are achieved by extracting the local extreme point sequences of the data point set and the precursor signal sequence, and comparing the degree of matching between the two extreme point sequences in terms of position and amplitude. Adaptive weighting coefficients are assigned to the distributed moment distance, dynamic time warping distance, and waveform structure distance, respectively; The weighted distribution moment distance, the dynamic time warping distance, and the waveform structure distance are fused to obtain the comprehensive distance.

[0010] Preferably, the source analysis and knowledge matching module summarizes recurring typical behavioral patterns based on the evolutionary path, including the following steps: Extract multiple fixed-length sub-path segments from the evolution path; Cluster all sub-path segments to form several sub-path segment clusters; Select the most representative central sub-path segment from each sub-path segment cluster; Each central sub-path segment is smoothed and generalized to eliminate random fluctuations while retaining its core trend of change. Define the processed central sub-path segment as a typical behavior pattern; A feature label is created for each typical behavior pattern, and the feature label is used to describe the common operating state of the puncture system corresponding to the typical behavior pattern.

[0011] Preferably, the clustering operation on all sub-path segments includes the following steps: Initialize a clustering model containing a preset number of centroids; Calculate the distance from each sub-path segment to all centroids; Based on the nearest neighbor principle, each sub-path segment is assigned to a cluster represented by a centroid; After the allocation is completed, the average value of all sub-path segments in each cluster is recalculated, and the average value is updated as the new centroid of the cluster; Iteratively execute the steps of allocating sub-path segments and updating centroids until the change in the position of the centroid is less than the convergence threshold or the maximum number of iterations is reached. The final cluster partitioning result and the corresponding centroid are used as the output of the sub-path segment cluster.

[0012] Preferably, the diagnosis and control instruction generation module generates a set of control instructions for potential operational deviations based on the descriptive diagnostic fragment, including the following steps: The descriptive diagnostic fragments are analyzed to identify the types of abnormal states, severity levels, and possible directions of evolution contained therein; Query the preset control strategy mapping table, which stores the correspondence between different abnormal state types, severity levels and recommended control actions; Based on the identified abnormal state type and severity level, the basic control instruction set is retrieved from the control strategy mapping table; Based on the possible evolution directions, the basic control instruction set is optimized and sorted in terms of timing; For each instruction in the optimized and sorted control instruction set, add the triggering condition and priority parameters for executing the instruction; The set of control instructions with attached trigger conditions and priority parameters is defined as the final set of control instructions.

[0013] Preferably, the step of optimizing the timing of the basic control instruction set in conjunction with the possible evolution directions includes the following steps: Obtain the system state at several consecutive future time points predicted by the possible evolution directions; For each instruction in the basic control instruction set, simulate its impact on the system state when it is executed at the current and future points in time; Evaluate the expected benefits and potential risks of executing each instruction at different points in time; With the goal of maximizing overall benefits and minimizing overall risks, an instruction timing optimization model is constructed. Solve the instruction timing optimization model to obtain the optimal suggested execution time for each control instruction; Based on the optimal recommended execution time, all control instructions are rearranged to form an optimized instruction sequence.

[0014] Preferably, the diagnosis and control instruction generation module matches the source signal features and their time-related events with a preset puncture system working logic knowledge base, including the following steps: Retrieve all logical rule entries containing the source signal characteristics from the puncture system's working logic knowledge base; From the retrieved logical rule entries, a subset that is simultaneously related to the time-related events is further filtered out; For the selected subset of logical rule entries, verify whether the source signal features and their time-related events satisfy the condition combination specified by the logical rule entries; If the condition is met, the logical rule entry is marked as a successfully matched entry. Extract the conclusion portion of all successfully matched entries; The conclusions are integrated and redundancy is removed to form a logical explanation of the current abnormal situation. This logical explanation constitutes the core content of the descriptive diagnostic segment.

[0015] Preferably, verifying whether the source signal features and their time-related events satisfy the condition combination specified by the logical rule entries includes the following steps: The condition part of the parsed logical rule entry is composed of multiple atomic conditions connected by logical operators; Each atomic condition is compared with the specific attributes of the source signal characteristics or its time-related events; Determine whether the condition for each atom is satisfied and record the Boolean truth value; Perform combination operations on the Boolean truth values ​​of all atomic conditions according to the logical operators defined in the logical rule entries; If the final result of the combination operation is true, then it is determined that the source signal features and their time-related events satisfy the combination of conditions specified in the logical rule entries.

[0016] Compared with the prior art, the beneficial effects of the present invention are: A time- and event-dimensional mapping network is established for the preprocessed effective signal components, and this network is used to track and record the complete evolution path of the signal within a continuous time window. Discrete time-point data are concatenated into a structured behavioral trajectory with contextual relevance. Unlike conventional techniques that isolate the instantaneous characteristics of signals, this approach enables the system to identify the dynamic processes and trends of signal changes. The effect is that the system can understand the puncture operation as a continuous process consisting of a series of sub-events, identify specific composite behavioral patterns composed of multiple steps, and keenly capture subtle early warning signals that may lead to final deviations.

[0017] A source tracing analysis is performed on the marked set of abnormal data points to trace their source signal characteristics and corresponding time-related events. This "feature-event" combination is then matched against a pre-defined knowledge base of the puncture system's operational logic, generating descriptive diagnostic fragments based on the matching results. This deeply binds abnormal phenomena to their specific operational causes and domain knowledge, enabling logical reasoning and explanation of the abnormalities' causes. The output diagnostic fragments clearly point to specific operational scenarios or instrument states, making the subsequently generated set of control instructions highly causal and actionable. Simultaneously, this process continuously matches real-world cases with the knowledge base, driving the continuous evolution and precision of the system's diagnostic logic. Attached Figure Description

[0018] Figure 1 This is a timing diagram of the puncture system for treating cardiovascular diseases according to the present invention; Figure 2 A flowchart for source tracing analysis of a set of data points; Figure 3 A flowchart for calculating morphological similarity; Figure 4 A comparison diagram of the signal evolution path during the monitoring phase of the puncture system. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Please see Figure 1This invention provides a puncture system for cardiovascular disease treatment. During operation, the system first receives real-time monitoring data of the puncture process from external sensing devices via a data receiving and preprocessing module. This data may include various signals from imaging devices, pressure sensors, and position sensors. This module performs multi-level noise suppression processing on the received raw data, using methods such as wavelet transform and adaptive filtering to progressively filter out power frequency interference, electromyographic noise, and random noise, thereby extracting effective signal components that clearly characterize the system's operational state, including needle movement, tissue interaction, and hemodynamic changes. Subsequently, a signal mapping and path tracing module establishes a mapping network integrating a time axis and an event axis for these purified effective signal components. This network associates each signal change with a specific timestamp and the event that may trigger the change (such as "beginning needle insertion" or "passing through the blood vessel wall"). The network continuously tracks and records the continuous change trajectory of the effective signal components within a sliding time window, i.e., its evolution path. The behavior pattern summarization and deviation marking module, based on accumulated historical evolution path data, uses pattern recognition algorithms to summarize the recurring behavior patterns of the system under normal or typical abnormal operating conditions. It compares the real-time evolution path acquired within the current time window with these typical behavior patterns, calculates the degree of deviation, and marks data points whose deviation values ​​exceed a preset deviation threshold, forming a set of data points to be analyzed. The source analysis and knowledge matching module is responsible for in-depth analysis of this data point set. Its core task is to trace the source signal characteristics that led to these abnormal data and their temporally related operations or physiological events. It then matches these traced characteristics and events with a preset knowledge base containing the puncture system's working logic, common fault modes, and physiological response patterns. Finally, the diagnosis and control instruction generation module generates a descriptive diagnostic segment based on the knowledge base matching results, describing the current operating status of the puncture system, the cause of the abnormality, and potential risks. Based on this diagnostic segment, it logically derives a series of control instructions to correct or prevent further deterioration of the operating status, thus forming a set of control instructions that can be output to the actuator or prompted to the operator.

[0021] Example 1: See Figure 2 This embodiment details the specific execution steps of the source tracing analysis, performing source tracing analysis on the set of data points marked by the behavior pattern induction and deviation marking module, including: Analyze the temporal and amplitude distribution characteristics of this data point set, such as whether its outliers are concentrated or sporadic, and what the statistical characteristics of amplitude deviation are.

[0022] Retrieve historical anomaly event records with similar distribution characteristics from the historical monitoring database. Extract precursor signal sequences that appeared before the final failure or anomaly from these retrieved historical anomaly event records. Calculate the morphological similarity between the current data point set and each extracted historical precursor signal sequence. Based on the calculated morphological similarity, select the target precursor signal sequence that is most similar to the current situation and has the strongest reference value. Analyze the complete failure event chain corresponding to this target precursor signal sequence in its historical records. From this historical failure event chain, extract the core trigger signal combination believed to have led to the final problem.

[0023] In the real-time monitoring data stream of the ongoing puncture procedure, a reverse search is performed to trace the occurrence of this core triggering signal combination. If traces of this core triggering signal combination are found, its specific form in the current monitoring data and its precise occurrence time are defined together as the source signal characteristics. Simultaneously, device operation events or physiological parameter events that are strongly correlated in time with the current form of the core triggering signal combination, such as a sudden increase in catheter torque or a sudden drop in patient blood pressure recorded within seconds before or after the occurrence of the triggering signal, are defined as time-related events.

[0024] In practice, the source analysis and knowledge matching module performs source analysis on the data point set. Real-time monitoring data during the puncture process includes echo intensity signals from the intravascular ultrasound catheter and intraluminal pressure signals from the pressure sensor. The behavior pattern induction and deviation labeling module identifies a set of abnormal pressure peak data points occurring within three consecutive cardiac cycles. Analysis of the temporal and amplitude distribution characteristics of the data point set shows that the peak intervals are asynchronous with the cardiac cycle, and the amplitudes exceed three standard deviations of the normal systolic blood pressure fluctuation range. The system retrieves abnormal event records with similar distribution characteristics from the historical monitoring database, using the search criteria of "peak intervals between 300 and 500 milliseconds and amplitudes exceeding 150% of the baseline." Five historical abnormal event records are returned. A precursor signal sequence is extracted from these five records; each precursor signal sequence contains multi-dimensional signal fragments within a 10-second time window preceding the abnormal pressure peak.

[0025] The system calculates the morphological similarity between the data point set and each precursor signal sequence. A dynamic time warping algorithm is used to calculate the cumulative distance after time alignment and normalizes it to a similarity score between 0 and 1. Based on morphological similarity, the most relevant target precursor signal sequence is selected. The scores of five historical precursor signal sequences are 0.35, 0.91, 0.60, 0.45, and 0.72, respectively. The precursor signal sequence with a score of 0.91 is identified as the target precursor signal sequence. The historical fault event chain corresponding to the target precursor signal sequence is analyzed. Historical records indicate that a contrast agent extravasation radiographic event occurred after the precursor signal sequence. The core trigger signal combination is extracted from the historical fault event chain and defined as "echo intensity signal attenuation exceeding 20 dB within 2 seconds accompanied by micro-oscillations in pressure signal with a frequency higher than 50 Hz".

[0026] The system searched for traces of the core trigger signal combination in the real-time monitoring data of the current puncture procedure. At the timestamp T-12 seconds, the system detected the echo intensity signal attenuating from 65 dB to 42 dB, and at the same time, detected a 52 Hz oscillation component in the pressure signal spectrum. The traces of the core trigger signal combination were found, and its current form and precise occurrence time were defined as the source signal characteristic, recorded as "Echo Attenuation Event @ T-12 seconds: Attenuation value 23 dB, duration 1.8 seconds; micro-oscillation frequency 52 Hz, average amplitude 0.8 mmHg". Equipment operation events or physiological parameter events that are strongly correlated with the current form of the core trigger signal combination in time were defined as time-related events. Within the time window [T-13 seconds, T-11 seconds], the system log recorded a "rapid guidewire advance" operation, which was defined as a time-related event.

[0027] In some embodiments, analyzing the temporal and amplitude distribution characteristics of a data point set includes calculating the peak interval distribution histogram and the probability density function of the amplitude of the data point set. The temporal distribution characteristics are characterized by a peak interval coefficient of variation greater than 0.3, and the amplitude distribution characteristics are characterized by a kurtosis greater than 5. When retrieving abnormal event records with similar distribution characteristics from the historical monitoring database, the system uses a matching algorithm based on distribution moments. The mean, variance, skewness, and kurtosis of the data point set are used to form a feature vector, which is then used to calculate the Euclidean distance between this feature vector and the feature vector of the historical records. Historical records with a distance less than a threshold are retrieved. The extracted precursor signal sequence is a fixed-length time series data truncated from the time of the abnormal event in the historical records.

[0028] It is understandable that calculating the morphological similarity between the data point set and each precursor signal sequence is a key step in source tracing analysis. The morphological similarity score integrates signal shape, temporal phase, and statistical characteristics. Based on morphological similarity, the target precursor signal sequence with the strongest reference value is selected using a ranking method, choosing the precursor signal sequence with the highest score, or a clustering method is used to select the central sequence of a representative cluster. Analyzing the historical fault event chain corresponding to the target precursor signal sequence involves retrieving the complete event log associated with that precursor signal sequence number. Extracting the core trigger signal combination from the historical fault event chain is achieved through a causal analysis model. The model identifies the initial signal change pattern in the event chain most likely to trigger a series of subsequent events.

[0029] Optionally, a sliding time window matching method can be used to search for traces of the core trigger signal combination in the real-time monitoring data of the current puncture process. The core trigger signal combination is used as a template, and convolution or correlation calculations are performed on the real-time data stream. After finding traces of the core trigger signal combination, defining the source signal characteristics requires recording the specific value, morphological description, and timestamp of each signal component within the combination in the current data. Defining time-related events requires setting a time-related threshold, including all logged operation events and alarm events whose absolute difference from the source signal characteristic timestamp is less than this threshold into the related event set.

[0030] In some embodiments, if no trace of the core triggering signal combination is found in the current real-time monitoring data of the puncture procedure, the source tracing analysis process will backtrack to re-screening the target prodromal signal sequence or expanding the search scope of the historical database. When defining time-related events, the criteria for judging strong correlation may include not only time proximity but also correlation between event types. For example, the correlation weight between the "guidewire operation" event and the "pressure fluctuation" signal is higher than that of the "bedside monitor alarm" event. The final output of the source signal characteristics and their time-related events is transmitted to the knowledge matching module in the form of structured data objects.

[0031] Example 2: See Figure 3When calculating the morphological similarity between the data point set and each precursor signal sequence, the data point set and each precursor signal sequence are first normalized to eliminate absolute numerical differences caused by different signal sources or dimensions. For the normalized data point set, its statistical distribution moments, including mean, variance, skewness, and kurtosis, are calculated. Similarly, for each normalized precursor signal sequence, its corresponding statistical distribution moments are calculated. To comprehensively assess similarity, a similarity assessment framework containing multiple distance metrics needs to be constructed. This framework must include at least distribution moment distance, dynamic time warping distance, and waveform structure distance. Using this similarity assessment framework, the comprehensive distance between the data point set and each precursor signal sequence is calculated one by one. This comprehensive distance is then converted into an intuitive morphological similarity score using a preset transformation function. Finally, a correspondence is established between the morphological similarity score and the index of each precursor signal sequence. When constructing the similarity assessment framework containing multiple distance metrics, the calculation rules for each distance metric need to be clearly defined. The calculation rules for the distributed moment distance are defined as follows: The distributed moment distance is calculated by comparing the differences between the data point set and the precursor signal sequence in four statistical moments (mean, variance, skewness, and kurtosis) and calculating their Euclidean or Mahalanobis distances. The calculation rules for the dynamic time warping distance are defined as follows: The optimal nonlinear alignment path between the time series of the data point set and the time series of the precursor signal sequence is found using a dynamic programming algorithm, and the cumulative distance between corresponding points after alignment along this optimal path is calculated. The calculation rules for the waveform structure distance are defined as follows: The local extremum sequences of the data point set and the precursor signal sequence are extracted, and the degree of matching between the two extremum sequences in terms of the timing and magnitude of the extremum points is compared. Adaptive weighting coefficients are assigned to the distributed moment distance, dynamic time warping distance, and waveform structure distance, and these coefficients can be dynamically adjusted according to the signal type or analysis stage. Finally, the weighted distributed moment distance, dynamic time warping distance, and waveform structure distance are fused, for example, through weighted summation or rule-based comprehensive judgment, to obtain a comprehensive distance used to measure the overall difference.

[0032] In the specific implementation, the source analysis and knowledge matching module calculates the morphological similarity between the data point set and each precursor signal sequence. The data point set contains five consecutive abnormal impedance values, with original values ​​of [120, 118, 125, 200, 210] ohms. In the specific implementation, the data point set and each precursor signal sequence are normalized to eliminate absolute dimensional differences. The maximum-minimum normalization method is used to linearly transform the values ​​of the data point set to the [0, 1] interval, resulting in a normalized data point set of [0.0, 0.0, 0.08, 0.89, 1.0]. In the specific implementation, the statistical distribution moments of the normalized data point set are calculated, yielding a mean of 0.394, a variance of 0.187, a skewness of 0.632, and a kurtosis of -1.085. In practice, the statistical distribution moments are calculated for each normalized precursor signal sequence. For example, if the normalized value of a historical precursor signal sequence is [0.1, 0.15, 0.12, 0.9, 0.95], the calculated statistical distribution moments are: mean 0.444, variance 0.132, skewness 0.815, and kurtosis -0.742.

[0033] In the specific implementation, a similarity assessment framework incorporating multiple distance metrics is constructed, including distribution moment distance, dynamic time warping distance, and waveform structure distance. The framework is used to calculate the comprehensive distance between the data point set and each precursor signal sequence. The calculated comprehensive distance is then converted into a morphological similarity score using a negative exponential mapping function. The morphological similarity score is set between 0 and 1, with higher scores indicating greater morphological similarity. Furthermore, an indexed correspondence is established between the morphological similarity score and the precursor signal sequence, storing the score results as key-value pairs, such as {"Sequence ID_001": 0.85, "Sequence ID_002": 0.72}.

[0034] In some embodiments, the calculation rule for the distributed moment distance is defined by comparing the differences in mean, variance, skewness, and kurtosis between the data point set and the precursor signal sequence. The distributed moment distance uses Euclidean distance to calculate the difference between the four statistical moment eigenvectors. In a specific implementation, the calculation rule for the dynamic time warping distance is defined by finding the optimal nonlinear alignment path between the data point set time series and the precursor signal sequence time series and calculating the cumulative distance under the optimal nonlinear alignment path. The cumulative distance is the sum of the absolute differences between all corresponding points of the two sequences after dynamic time warping alignment. In a specific implementation, the calculation rule for the waveform structure distance is defined by extracting the local extremum sequences of the data point set and the precursor signal sequence and comparing the degree of matching between the two extremum sequence sequences in terms of position and amplitude. The degree of matching is jointly determined by the penalty for extremum point position offset and the weighted sum of amplitude differences. In practice, adaptive weighting coefficients are assigned to the distributed moment distance, dynamic time warping distance, and waveform structure distance. These coefficients are dynamically adjusted based on the signal type. For slowly changing pressure signals, the distributed moment distance has a higher weighting coefficient, while for rapidly changing electrophysiological signals, the dynamic time warping distance has a higher weighting coefficient. The weighted distributed moment distance, dynamic time warping distance, and waveform structure distance are then fused to obtain the comprehensive distance. The comprehensive distance is calculated using a linear weighted sum formula.

[0035] In practical implementation, the formula for the linear weighted sum of the comprehensive distances is expressed as:

[0036] Among them, characters Represents the calculated composite distance, character Represents the adaptive weighting coefficients assigned to the distance of the distributed moments, character Represents the calculated distribution moment distance, character This represents the adaptive weighting coefficients assigned to the dynamic time-warped distance, character Represents the calculated dynamic time-warped distance, character Represents the adaptive weighting coefficients assigned to the waveform structure distance, character This represents the calculated waveform structure distance.

[0037] It is understandable that normalizing the data point set and each precursor signal sequence is the foundation for subsequent distance calculations. Normalization eliminates differences in amplitude reference and dimensions of the original signals, making morphological comparisons between different sensors or different physiological parameters possible. Calculating the statistical distribution moments of the normalized data can capture the overall digital characteristics of the signal waveform: the mean reflects the average level, the variance reflects the fluctuation range, the skewness reflects asymmetry, and the kurtosis reflects the sharpness.

[0038] Optionally, a similarity assessment framework incorporating multiple distance metrics is constructed because a single distance metric may not fully capture the similarity of complex physiological signals. Distribution moment distance focuses on overall statistical characteristics, dynamic time warping distance focuses on morphological similarity under temporal curvature, and waveform structure distance focuses on the correspondence of local extrema. The process of calculating the comprehensive distance using the similarity assessment framework is an independent cyclic or vectorized calculation process for each precursor signal sequence. The comprehensive distance is converted into a morphological similarity score through a mapping function, such as using a formula. or Where S represents the morphological similarity score and D represents the overall distance.

[0039] In some embodiments, normalization can also employ Z-score standardization to transform the original data into a distribution with a mean of 0 and a standard deviation of 1. The statistical distribution moments can be calculated using the formula for higher-order central moments. The dynamic time-warped distance can be calculated using a dynamic programming algorithm, where the cumulative distance path must satisfy boundary conditions, monotonicity, and continuity constraints. Extracting local extremum sequences can be achieved by finding points where the first derivative of the signal is zero and the sign of the second derivative changes. The allocation of adaptive weight coefficients can be adjusted online based on the signal-to-noise ratio or historical matching accuracy.

[0040] Assigning adaptive weights to the distribution moment distance, dynamic time warping distance, and waveform structure distance demonstrates the flexibility of the evaluation framework. These weights can be determined through pre-training or a rule base. The final quantitative step in similarity assessment is to fuse the weighted distances to obtain the comprehensive distance; a smaller comprehensive distance value indicates greater similarity between the two sequences. Establishing an index correspondence between morphological similarity scores and precursor signal sequences facilitates rapid retrieval and referencing of the highest-scoring historical sequences.

[0041] Example 3: Based on the evolutionary path, recurring typical behavioral patterns are summarized, and multiple fixed-length sub-path segments are extracted from the long-term recorded evolutionary path of the system. All acquired sub-path segments are clustered to form several sub-path segment clusters with high internal similarity and significant differences between each other. From each sub-path segment cluster, a central sub-path segment with the smallest average distance to all other segments within the cluster is selected as the representative of that cluster. Each selected central sub-path segment undergoes smoothing and generalization processing, such as using moving averages or spline fitting, to eliminate random fluctuation noise components in the segment and retain its core and most stable trend. The smoothed and generalized central sub-path segment is defined as a typical behavioral pattern. A feature label is created for each defined typical behavioral pattern, which describes the common operating state of the puncture system corresponding to this behavioral pattern, such as "smooth advancement," "impact vibration," or "wall-penetrating rebound." When performing clustering operations on all sub-path segments, a clustering model containing a preset number of centroids, such as a K-means clustering model, is initialized. Calculate the distance from each sub-path segment to all centroids. Based on the nearest neighbor principle, assign each sub-path segment to a cluster represented by its nearest centroid. After all sub-path segments are assigned, recalculate the average value of all sub-path segments in each cluster and update this average value as the new centroid for that cluster. Iteratively execute the steps of assigning sub-path segments and updating centroids until the positional change of all centroids is less than a preset convergence threshold or the maximum number of iterations is reached. Output the final cluster partitioning result and the final centroid for each cluster as the sub-path segment clusters.

[0042] In practical implementation, the behavior pattern induction and deviation labeling module summarizes recurring typical behavior patterns based on the evolution path. The evolution path is a multi-dimensional time series with 1000 sampling points obtained from the signal mapping and path tracing module. This series describes the joint change trajectory of the puncture needle tip pressure and displacement signal within the last 10 seconds. From the 1000-segment evolution path, multiple sub-path segments with a fixed length of 100 sampling points are extracted using a sliding window method. The overlap length between adjacent sub-path segments is 50 sampling points, resulting in a total of 19 sub-path segments. All 19 sub-path segments are clustered, with a preset cluster size of 4, forming 4 sub-path segment clusters. From each sub-path segment cluster, the most representative central sub-path segment is selected. The selection criterion is to calculate the average distance from each sub-path segment within the cluster to all other sub-path segments within the same cluster, and the sub-path segment with the smallest average distance is defined as the central sub-path segment. Each central sub-path segment undergoes smoothing and generalization processing, and a local weighted regression method is used to eliminate random fluctuations in the sub-path segments while retaining their core change trends. The four processed central sub-path segments were defined as four typical behavior patterns: Pattern A, Pattern B, Pattern C, and Pattern D. A feature label was created for each typical behavior pattern to describe the common operating state of the puncture system corresponding to that pattern. The feature label for Pattern A was "smooth advancement", the feature label for Pattern B was "increased tissue resistance", the feature label for Pattern C was "slight shaking", and the feature label for Pattern D was "vessel wall contact".

[0043] In the specific implementation, clustering is performed on all sub-path segments. A clustering model with a preset number of centroids (4) is initialized, and 4 centroids are randomly selected from 19 sub-path segments as initial centroids. The distance from each sub-path segment to all centroids is calculated using the Euclidean distance formula, which measures the square root of the sum of squared differences between corresponding points in two sequences of length 100. Based on the nearest neighbor principle, each sub-path segment is assigned to a cluster represented by its nearest centroid. After assignment, the average value of all sub-path segments in each cluster is recalculated by taking the arithmetic mean of the values ​​of all sub-path segments within the cluster at each time point, generating a new sequence of length 100. The calculated average value sequence is then used to update the new centroid of the corresponding cluster. The steps of assigning sub-path segments and updating centroids are iteratively executed until the change in the position of the centroid is less than the convergence threshold or the maximum number of iterations is reached. The convergence threshold is set to 0.001, and the maximum number of iterations is set to 300. The final clustering result and the corresponding centroids are used as the output of the sub-path segment clusters. The output includes the label number of the cluster to which each sub-path segment belongs and four final centroid sequences.

[0044] In some embodiments, extracting multiple fixed-length sub-path segments from the evolutionary path can employ non-overlapping fixed windows or variable-length windows to adapt to patterns of different durations. Clustering all sub-path segments can be performed using K-means clustering, hierarchical clustering, or density clustering. Selecting the most representative central sub-path segment from each cluster can be done by directly using the calculated cluster centroid, or by selecting the actual segment closest to the centroid through voting or density estimation. Smoothing and generalization processing for each central sub-path segment can be performed using moving average filtering, Gaussian smoothing, or spline interpolation.

[0045] It is understandable that extracting multiple fixed-length sub-path segments from the evolutionary path is the process of transforming a continuous time series into standard data units suitable for pattern analysis. Clustering all sub-path segments is a method based on unsupervised learning to discover the inherent structure of the data, grouping similar morphological change trajectories into one category. Selecting the most representative central sub-path segment from each sub-path segment cluster is a crucial step in extracting the core information of the cluster; the central sub-path segment summarizes the entire cluster. Smoothing and generalization processing of each central sub-path segment aims to remove random noise and individual operational differences during data acquisition, thereby extracting stable and repeatable signal change patterns.

[0046] In practice, the clustering operation recalculates the average value of all sub-path segments in each cluster to update the centroid. The calculation formula is as follows:

[0047] Among them, characters This represents the new centroid vector of the k-th cluster after the update, where the character... Represents the set of all sub-path fragments in the k-th cluster, characters Represents a set Number of neutron path segments, symbol Represents a set Each sub-path segment vector in To perform summation, characters Indicates belonging to a cluster The vector of the i-th sub-path segment.

[0048] Optionally, when initializing the clustering model, the preset number can be determined in advance based on historical experience or through elbow rule analysis. The distance from each sub-path segment to all centroids can be calculated using other metrics such as Manhattan distance or cosine distance. When assigning sub-path segments according to the nearest neighbor principle, if a sub-path segment has the same distance to multiple centroids, it is randomly assigned or assigned according to the cluster size. Recalculating the mean of each cluster as the new centroid is a standard step in the K-means algorithm, ensuring that the centroid is the mean center of the points within the cluster. Iteratively executing the assignment and update steps until the centroids stabilize is a necessary process for algorithm convergence, and the convergence threshold controls the accuracy of the results.

[0049] In some embodiments, the input sub-path segments for clustering operations can be either the original signal values ​​or feature vectors after feature extraction. The smoothing and generalization processing performed after selecting the central sub-path segment can adaptively adjust its smoothing intensity parameter according to the signal's noise level. After defining typical behavior patterns, the system assigns a unique identifier to each pattern and stores it in a pattern library. The feature labels established for each typical behavior pattern can be text descriptions or classification codes for fast retrieval. The typical behavior pattern library is updated periodically as the system collects new evolutionary paths; this update process can trigger new clustering calculations.

[0050] When performing clustering operations on all sub-path segments to summarize typical behavioral patterns, a standardized data analysis process can be used. The signal evolution path is preprocessed by first acquiring multi-dimensional evolution path data output from the signal mapping and path tracing module, then extracting sub-path segments of fixed length. For multi-dimensional sub-path segments, they need to be converted into one-dimensional feature vectors to unify the analysis dimensions. The clustering model is initialized by pre-setting the number of clusters based on the number of categories in common operating states of the puncture system, while also setting the maximum number of iterations and convergence threshold to ensure the stability and accuracy of the clustering process. After model initialization, the processed sub-path segment feature vectors are input into the model. The model automatically performs data fitting, first calculating the distance from each sub-path segment to its initial centroid, then performing the first round of cluster allocation based on the nearest neighbor principle. The cluster centroids are then iteratively updated repeatedly until the centroid position changes meet the convergence requirements or reach the iteration limit. Finally, the cluster label for each sub-path segment and the final centroid of each cluster are output. After clustering is completed, for each cluster, the sub-path segment closest to the corresponding centroid needs to be selected as the representative segment of the cluster. In this way, the core signal change characteristics of each cluster can be extracted, which can achieve accurate classification of signal evolution trajectory.

[0051] Example 4: The diagnosis and control instruction generation module generates a set of control instructions for potential operational deviations based on descriptive diagnostic fragments. First, it parses the descriptive diagnostic fragments to identify the types of abnormal states, severity levels, and possible evolution directions predicted based on the evolution path. It then queries a pre-set control strategy mapping table, which stores the correspondence between different abnormal state types, severity levels, and a series of recommended control actions. Based on the identified abnormal state types and severity levels, it retrieves the corresponding basic control instruction set from the control strategy mapping table. Combining the possible evolution direction predictions given in the diagnostic fragments, it performs temporal optimization sorting of the basic control instruction set. For each instruction in the optimized control instruction set, it adds the triggering conditions and priority parameters required to execute the instruction. The control instruction set with the added triggering conditions and priority parameters is defined as the final output control instruction set. When performing temporal optimization sorting of the basic control instruction set in conjunction with possible evolution directions, it obtains the system state at several consecutive future time points predicted by the possible evolution directions. For each instruction in the basic control instruction set, it simulates the expected impact on the system state when executed at the current time point and at each predicted future time point. Evaluate the expected benefits and potential risks of executing each instruction at different time points. Construct an instruction timing optimization model with the optimization objective of maximizing overall benefits and minimizing overall risks. Solve this model to obtain the optimal suggested execution time for each control instruction. Based on the calculated optimal suggested execution times, rearrange all control instructions to form an optimized instruction sequence.

[0052] In practical implementation, the diagnosis and control instruction generation module generates a set of control instructions for potential operational deviations based on descriptive diagnostic fragments. The content of the descriptive diagnostic fragments is "intermittent inaccuracy of pressure sensor signal, moderate deviation, predicted deviation may persist and affect pressure guidance accuracy." The module analyzes the descriptive diagnostic fragments to identify the abnormal state type, severity level, and possible evolution direction. The analyzed abnormal state type is "signal inaccuracy," the severity level is "moderate," and the possible evolution direction is "persistent deviation affecting pressure guidance accuracy." A preset control strategy mapping table is queried; this table is stored in a relational database. Based on the identified abnormal state type "signal inaccuracy" and severity level "moderate," the basic control instruction set is retrieved from the control strategy mapping table. Three basic control instructions are retrieved: "Instruction A: Initiate backup pressure sensor calibration process," "Instruction B: Switch to image navigation priority mode," and "Instruction C: Prompt operator to manually verify pressure value." Combining the predicted evolution direction of "persistent deviation affecting pressure guidance accuracy" in the descriptive diagnostic fragments, the basic control instruction set is optimized and ordered in a temporal sequence. Each instruction in the optimized and ordered control instruction set is appended with a trigger condition and priority parameter. The control instruction set with the appended trigger conditions and priority parameters is defined as the final control instruction set, which is then output to the system control unit.

[0053] In practical implementation, the basic control command set is optimized and ordered temporally based on possible evolution directions to obtain the system state at several consecutive future time points predicted by the possible evolution directions. The prediction model outputs the system pressure reading confidence levels for the next five time points (T+1 to T+5) as [0.7, 0.6, 0.5, 0.4, 0.3]. For each command in the basic control command set, its impact on the system state when executed at the current and future time points is simulated. The simulation shows that executing "Command A: Initiate backup pressure sensor calibration process" at the current time point will increase the confidence level of pressure readings at all subsequent time points to above 0.9. The expected benefits and potential risks of executing each command at different time points are evaluated. For example, the expected benefit of executing Command A at time T+1 is rapid restoration of measurement reliability, while the potential risk is a brief data interruption during calibration. An optimized command timing model is constructed with the goal of maximizing overall benefits and minimizing overall risks. The optimal suggested execution time for each control instruction is obtained by solving the instruction timing optimization model. The result is that instruction A is executed at time T+1, instruction B at time T+2, and instruction C at time T+3. Based on the optimal suggested execution time, all control instructions are rearranged to form an optimized instruction sequence. The final sequence is [instruction A@T+1, instruction B@T+2, instruction C@T+3].

[0054] Table: Fragment of the Regulation Strategy Mapping Table

[0055] In some embodiments, parsing descriptive diagnostic fragments can employ named entity recognition technology in natural language processing to extract keywords, or it can parse predefined structured diagnostic data objects. Querying the pre-defined control strategy mapping table can be achieved using a database query language, with the abnormal state type and severity level as joint query conditions. The retrieved basic control instruction set may contain multiple instructions, which may have logical dependencies or mutual exclusion relationships. When performing time-series optimization and sorting based on possible evolution directions, the evolution direction can be quantified as a predicted probability distribution of the system state at a series of future time points.

[0056] It is understandable that parsing descriptive diagnostic fragments is the logical starting point for generating targeted control instructions, and clearly defined abnormal states and severity are the foundation for retrieving control strategies. Querying a pre-built control strategy mapping table establishes a mapping relationship from problem diagnosis to solutions; this mapping table is pre-constructed based on domain knowledge. Retrieving the basic control instruction set based on the identified type and level ensures that the control actions are basically matched with the current problem. Optimizing the temporal order of the basic control instruction set introduces the prediction of fault development trends, transforming the control strategy from a static response to dynamic programming.

[0057] In practical implementation, the constructed instruction timing optimization model aims to maximize the overall net benefit, and its objective function can be expressed as:

[0058] Among them, characters Represents the set of all considered current and future time points, character This represents the set of all instructions in the basic control instruction set, character. Indication of instructions At the point of time Quantitative value of the expected benefits generated by the execution, character Indication of instructions At the point of time The quantification of potential risks generated during execution, in characters It is a binary decision variable, when the instruction... Arranged at a certain time The value is 1 during execution and 0 otherwise. Solving this model involves assigning a value of 1 to each instruction. Determine one Time point .

[0059] Optionally, the expected benefits and potential risks can be quantified by querying historical data on the effects of similar instruction executions. Benefits and risks can be defined as their impact on key performance indicators (such as accuracy, stability, and safety scores). When constructing the instruction timing optimization model, the objective function can be a weighted sum of the benefits and risks of each instruction at each time point. Constraints can include the order of instructions and resource constraints such as the limitation on the number of instructions that can be executed at any given time. The instruction timing optimization model can be solved using an integer programming solver or a heuristic search algorithm. When rearranging the instruction sequence according to the optimal suggested execution time point, the temporal feasibility of the sequence must be ensured. The accompanying triggering condition can be an absolute time point or a relative system state event (such as "when the stress confidence level is below 0.5"). The priority parameter is used to determine the execution order when multiple instruction triggering conditions are met simultaneously.

[0060] In some embodiments, the control strategy mapping table can be designed as a multi-layered structure: the first layer maps the type of abnormal state, the second layer maps the severity, and the third layer maps the specific evolution direction or environmental context. Instructions in the basic control instruction set can include metadata, such as expected execution time, required resources, and expected duration of effect; this metadata can be used for timing optimization. Simulating the impact of instruction execution on the system state can be based on a simplified system dynamics model or a state transition probability matrix. The solution to the instruction timing optimization model can be a deterministic execution plan or a decision tree with probabilistic branches. The final control instruction set can be encapsulated as a machine-executable script or transformed into a list of steps for operator confirmation and execution.

[0061] Example 5: The diagnosis and control instruction generation module matches the source signal characteristics and their time-related events with a preset puncture system working logic knowledge base. It retrieves all logical rule entries from the puncture system working logic knowledge base whose condition parts contain the source signal characteristics. From the retrieved logical rule entries, it further filters out subsets that are simultaneously related to time-related events. For each filtered subset of logical rule entries, it verifies whether the source signal characteristics and their time-related events satisfy the condition combination specified by the logical rule entry. If satisfied, the logical rule entry is marked as a successful match. The conclusion parts of all successfully matched entries are extracted. These conclusion parts are integrated and deredundanted to form a logical explanation of the current abnormal situation; this logical explanation constitutes the core content of the descriptive diagnostic segment. When verifying whether the source signal characteristics and their time-related events satisfy the condition combination specified by the logical rule entry, the condition part of the logical rule entry is parsed. This part is usually composed of multiple atomic conditions connected by logical operators such as "AND," "OR," and "NOT." Each atomic condition is compared with the specific attributes of the source signal characteristics or their time-related events, such as determining whether "signal amplitude is greater than a threshold" or "event type equals a certain operation" is true. Each atomic condition is checked for satisfaction, and its Boolean truth value is recorded. The Boolean truth values ​​of all atomic conditions are combined according to the logical operators defined in the logical rule entry. If the final result of the combination operation is true, then the source signal characteristics and their time-related events are determined to satisfy the condition combination specified in the logical rule entry.

[0062] In practical implementation, the diagnosis and control instruction generation module matches the source signal characteristics and their time-related events with a preset puncture system working logic knowledge base. The source signal characteristic is "echo intensity signal attenuation exceeding 20 dB within 2 seconds," and the time-related event is "catheter torque sensor recording a peak value exceeding a threshold." All logical rule entries containing the source signal characteristic "echo intensity signal attenuation exceeding 20 dB within 2 seconds" are retrieved from the puncture system working logic knowledge base, returning three relevant rule entries. From these three retrieved logical rule entries, a subset simultaneously related to the time-related event "catheter torque sensor recording a peak value exceeding a threshold" is further filtered, resulting in two logical rule entries. For these two selected logical rule entries, the source signal characteristic "echo intensity signal attenuation exceeding 20 dB within 2 seconds" and its time-related event "catheter torque sensor recording a peak value exceeding a threshold" are verified to ensure they meet the condition combination specified by the logical rule entry. The verification results show that the condition combination of the first logical rule entry is met, while the condition combination of the second logical rule entry is not met. The first logical rule entry that meets the conditions is marked as a successful match. The conclusion portion of the successful match entry is extracted, which is "suspected abnormal contact or tissue damage between the catheter tip and the vessel wall". The extracted conclusion portion "suspected abnormal contact or tissue damage between the catheter tip and the vessel wall" is integrated and redundancy is removed to form a logical explanation of the current abnormal situation. This logical explanation constitutes the core content of the descriptive diagnostic segment.

[0063] In practical implementation, the source signal characteristics and their time-related events are verified to meet the condition combinations specified in the logical rule entries. The condition part of the logical rule entries is analyzed. The condition part of the first logical rule entry is "(echo attenuation > 15 dB) AND (torque peak > threshold) AND (event time difference < 1 second)", which consists of three atomic conditions connected by the logical operator "AND". Each atomic condition is compared with the specific attributes of the source signal characteristics or their time-related events. The first atomic condition "echo attenuation > 15 dB" is compared with the specific attribute of the source signal characteristic "attenuation value 23 dB". It is determined whether each atomic condition is satisfied and the Boolean truth value is recorded. The result of the first atomic condition "echo attenuation > 15 dB" is true, and the Boolean truth value is recorded as True. According to the logical operators defined in the logical rule entries, the Boolean truth values ​​of all atomic conditions are combined. The Boolean truth values ​​of the three atomic conditions are True, True, and True, respectively. The final result is True after combining them using the logical operator "AND". If the final result of the combination operation is true, then it is determined that the source signal characteristics and their time-related events satisfy the combination of conditions specified in the logical rule entry.

[0064] In some embodiments, retrieving logical rule entries from the puncture system's operational logic knowledge base can be done using a feature-based keyword index query. The knowledge base is stored in a rule engine format, with each rule having a unique identifier and a list of preconditions. Further filtering to identify subsets simultaneously related to time-related events can be achieved by matching event type fields in the search results. When verifying the combination of conditions specified by a logical rule entry, the condition portion may contain various atomic condition types, such as numerical comparison, string matching, set inclusion, or time interval judgment. Comparing atomic conditions with specific attributes requires mapping and calculating the structured data of the source signal features and time-related events with the variables and thresholds defined in the atomic conditions.

[0065] It is understandable that retrieving all logical rule entries containing source signal features from the puncture system's working logic knowledge base is the entry point for knowledge matching, ensuring that all potentially relevant rules are considered. Further filtering to identify subsets simultaneously related to time-related events narrows the matching range, improving accuracy and efficiency. Verifying whether the source signal features and their time-related events satisfy the conditional combinations specified by the logical rule entries is the core step of knowledge reasoning, determining whether a rule can be activated. Marking logical rule entries that meet the conditions as successfully matched confirms the reasoning results. Extracting the conclusion portions of all successfully matched entries aims to collect the conclusions derived from all activated rules. Integrating and deduplicating the conclusion portions to form a logical explanation generates a concise, consistent, and comprehensive diagnostic description.

[0066] In practical implementation, the combination operation of the condition part of the logical rule entries can be formally expressed as:

[0067] Among them, characters Represents the final Boolean value of the entire condition section, symbol This indicates a logical AND operation on all child items; the character... This represents the total number of atomic conditions in the logical rule entry, represented by the character. Indicates the first The Boolean truth value after each atomic condition is determined. Only when all When both are true, Only when the condition is true will the logical rule be satisfied.

[0068] Optionally, parsing the condition portion of a logical rule entry requires identifying the logical operators (such as "AND", "OR", and "NOT") within the conditions and their nested combinations. Comparing each atomic condition with specific attributes of the source signal characteristics or their time-related events involves data type conversions and value comparison operations, such as converting a textual description of a "threshold" into a specific numerical value stored in a database. The Boolean truth value of each atomic condition can be stored in a temporary array or dictionary for subsequent combination operations. Combination operations using the logical operators defined in the logical rule entry must adhere to standard Boolean algebra precedence and associativity. Once the source signal characteristics and their time-related events satisfy the combination of conditions specified in the logical rule entry, that logical rule entry is considered valid evidence.

[0069] In some embodiments, source signal features may simultaneously match multiple logical rule entries. These rule entries may lead to the same or complementary conclusions, or they may lead to seemingly contradictory conclusions. In such cases, conclusion fusion based on rule confidence or priority is required. Time-related events may contain multiple sub-events. When filtering relevant subsets, it is necessary to check whether the conditional part of the logical rule entry references the type or attribute of these sub-events. When verifying condition combinations, if the atomic condition involves time correlation (e.g., "Event A occurs within 5 seconds of Event B"), the time difference between the source signal feature and the time-related event needs to be accurately calculated. Logical rule entries in the knowledge base can be dynamically updated to incorporate new medical knowledge or operational experience. The final descriptive diagnostic fragment core content will be passed to the next processing stage of the diagnosis and regulation instruction generation module for generating regulation instructions.

[0070] See Figure 4 In the signal evolution analysis during the puncture system monitoring phase, the temporal amplitude variation characteristics of normal signals, amplitude exceeding limits abnormal signals, and frequency deviation abnormal signals were visually presented. Specifically, the amplitude fluctuation range of the normal signal (blue curve) is concentrated in the [-1, 1] interval, reflecting the signal baseline characteristics during stable system operation; the amplitude exceeding limits abnormal signal (red curve) has an amplitude that is consistently above 1, and at some points approaches 3, significantly deviating from the normal range, corresponding to amplitude-related abnormal states during system operation; the amplitude fluctuation frequency of the frequency deviation abnormal signal (yellow curve) is significantly higher than that of the normal signal, and the density of local extreme points is greater, reflecting the abnormal deviation of the signal frequency characteristics. The differences in the evolution paths of the three types of signals on the time axis provide direct visual evidence for the signal mapping and path tracing module to identify typical behavioral patterns and mark deviation data points, which can serve as the basic observation data for subsequent source tracing analysis and diagnostic control.

[0071] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0072] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A puncture system for treating cardiovascular diseases, characterized in that, The system includes: The data receiving and preprocessing module is used to receive real-time monitoring data of the puncture process from external devices, and to perform multi-level noise suppression on the real-time monitoring data of the puncture process to extract effective signal components characterizing the system operating status. The signal mapping and path tracing module is used to establish a time and event dimension mapping network for the effective signal components, and to track and record the evolution path of the effective signal components within a continuous time window through the mapping network. The behavior pattern induction and deviation marking module is used to inductively identify recurring typical behavior patterns based on the evolution path, and to compare the effective signal components within the current time window with the typical behavior patterns to mark the set of data points that exceed a preset deviation threshold. The source analysis and knowledge matching module is used to perform source analysis on the data point set to trace the source signal characteristics and time-related events that generated the data point set, and to match the source signal characteristics and time-related events with a preset puncture system working logic knowledge base; The diagnostic and control instruction generation module is used to generate a descriptive diagnostic fragment on the operating status of the puncture system based on the matching results, and to generate a set of control instructions for potential deviations in the operating status based on the descriptive diagnostic fragment.

2. The puncture system for treating cardiovascular diseases according to claim 1, characterized in that, The process of performing source tracing analysis on the data point set includes the following steps: Analyze the temporal and amplitude distribution characteristics of the data point set; Retrieve records of abnormal events with similar distribution characteristics from the historical monitoring database; Extract the precursor signal sequence from the abnormal event record; Calculate the morphological similarity between the data point set and each precursor signal sequence; Based on the morphological similarity, the target precursor signal sequence with the strongest reference value is selected; Analyze the historical fault event chain corresponding to the target precursor signal sequence; Extract the core trigger signal combination from the aforementioned chain of historical failure events; Search for traces of the core trigger signal combination in the real-time monitoring data of the current puncture procedure; If traces of the occurrence of the core trigger signal combination are found, the current form of the core trigger signal combination and its precise occurrence time are defined as source signal characteristics. Device operation events or physiological parameter events that are strongly correlated in time with the current form of the core trigger signal are defined as time-related events.

3. The puncture system for treating cardiovascular diseases according to claim 2, characterized in that, The calculation of the morphological similarity between the data point set and each precursor signal sequence includes the following steps: The data point set and each precursor signal sequence are normalized to eliminate differences in absolute dimensions. For the normalized set of data points, calculate its statistical distribution moments; For each normalized precursor signal sequence, calculate its corresponding statistical distribution moments; Construct a similarity evaluation framework that includes multiple distance metrics, including at least distribution moment distance, dynamic time warping distance, and waveform structure distance; Using the aforementioned similarity assessment framework, the comprehensive distance between the data point set and each precursor signal sequence is calculated one by one; The comprehensive distance is converted into a morphological similarity score; Establish an index correspondence between morphological similarity scores and precursor signal sequences.

4. The puncture system for treating cardiovascular diseases according to claim 3, characterized in that, The construction of a similarity evaluation framework that includes multiple distance metrics includes the following steps: The calculation rules for the distribution moment distance are defined, which are achieved by comparing the differences between the data point set and the precursor signal sequence in terms of mean, variance, skewness, and kurtosis. A calculation rule for dynamic time warping distance is defined, which is achieved by finding the optimal nonlinear alignment path between the time series of the data point set and the time series of the precursor signal sequence, and calculating the cumulative distance under the optimal nonlinear alignment path; The calculation rules for waveform structure distance are defined. These rules are achieved by extracting the local extreme point sequences of the data point set and the precursor signal sequence, and comparing the degree of matching between the two extreme point sequences in terms of position and amplitude. Adaptive weighting coefficients are assigned to the distributed moment distance, dynamic time warping distance, and waveform structure distance, respectively; The weighted distribution moment distance, the dynamic time warping distance, and the waveform structure distance are fused to obtain the comprehensive distance.

5. The puncture system for treating cardiovascular diseases according to claim 1, characterized in that, The source analysis and knowledge matching module summarizes recurring typical behavioral patterns based on the evolutionary path, including the following steps: Extract multiple fixed-length sub-path segments from the evolution path; Cluster all sub-path segments to form several sub-path segment clusters; Select the most representative central sub-path segment from each sub-path segment cluster; Each central sub-path segment is smoothed and generalized to eliminate random fluctuations while retaining its core trend of change. Define the processed central sub-path segment as a typical behavior pattern; A feature label is created for each typical behavior pattern, and the feature label is used to describe the common operating state of the puncture system corresponding to the typical behavior pattern.

6. The puncture system for treating cardiovascular diseases according to claim 5, characterized in that, The clustering operation for all sub-path segments includes the following steps: Initialize a clustering model containing a preset number of centroids; Calculate the distance from each sub-path segment to all centroids; Based on the nearest neighbor principle, each sub-path segment is assigned to a cluster represented by a centroid; After the allocation is completed, the average value of all sub-path segments in each cluster is recalculated, and the average value is updated as the new centroid of the cluster; Iteratively execute the steps of allocating sub-path segments and updating centroids until the change in the position of the centroid is less than the convergence threshold or the maximum number of iterations is reached. The final cluster partitioning result and the corresponding centroid are used as the output of the sub-path segment cluster.

7. The puncture system for treating cardiovascular diseases according to claim 1, characterized in that, The diagnosis and control instruction generation module generates a set of control instructions for potential operational deviations based on the descriptive diagnostic fragment, including the following steps: The descriptive diagnostic fragments are analyzed to identify the types of abnormal states, severity levels, and possible directions of evolution contained therein; Query the preset control strategy mapping table, which stores the correspondence between different abnormal state types, severity levels and recommended control actions; Based on the identified abnormal state type and severity level, the basic control instruction set is retrieved from the control strategy mapping table; Based on the possible evolution directions, the basic control instruction set is optimized and sorted in terms of timing; For each instruction in the optimized and sorted control instruction set, add the triggering condition and priority parameters for executing the instruction; The set of control instructions with attached trigger conditions and priority parameters is defined as the final set of control instructions.

8. The puncture system for treating cardiovascular diseases according to claim 7, characterized in that, The step of optimizing the timing sequence of the basic control instruction set in conjunction with the possible evolution directions includes the following steps: Obtain the system state at several consecutive future time points predicted by the possible evolution directions; For each instruction in the basic control instruction set, simulate its impact on the system state when it is executed at the current and future points in time; Evaluate the expected benefits and potential risks of executing each instruction at different points in time; With the goal of maximizing overall benefits and minimizing overall risks, an instruction timing optimization model is constructed. Solve the instruction timing optimization model to obtain the optimal suggested execution time for each control instruction; Based on the optimal recommended execution time, all control instructions are rearranged to form an optimized instruction sequence.

9. The puncture system for treating cardiovascular diseases according to claim 1, characterized in that, The diagnostic and control instruction generation module matches the source signal characteristics and their time-related events with a preset puncture system working logic knowledge base, including the following steps: Retrieve all logical rule entries containing the source signal characteristics from the puncture system's working logic knowledge base; From the retrieved logical rule entries, a subset that is simultaneously related to the time-related events is further filtered out; For the selected subset of logical rule entries, verify whether the source signal features and their time-related events satisfy the condition combination specified by the logical rule entries; If the condition is met, the logical rule entry is marked as a successfully matched entry. Extract the conclusion portion of all successfully matched entries; The conclusions are integrated and redundancy is removed to form a logical explanation of the current abnormal situation. This logical explanation constitutes the core content of the descriptive diagnostic segment.

10. The puncture system for treating cardiovascular diseases according to claim 9, characterized in that, The verification of whether the source signal features and their time-related events satisfy the combination of conditions specified in the logical rule entries includes the following steps: The condition part of the parsed logical rule entry is composed of multiple atomic conditions connected by logical operators; Each atomic condition is compared with the specific attributes of the source signal characteristics or its time-related events; Determine whether the condition for each atom is satisfied and record the Boolean truth value; Perform combination operations on the Boolean truth values ​​of all atomic conditions according to the logical operators defined in the logical rule entries; If the final result of the combination operation is true, then it is determined that the source signal features and their time-related events satisfy the combination of conditions specified in the logical rule entries.

Images