Method for monitoring signal visualization applied to implantable medical monitoring sensors

Abstract

The application provides a monitoring signal visualization method applied to an implantable medical monitoring sensor, relates to the field of data processing, and comprises the following steps: after a target monitoring signal data stream and detection guide instructions corresponding to the target monitoring signal data stream are acquired, a monitoring signal segment in the target monitoring signal data stream is implicitly represented, and the detection guide instructions are implicitly represented; then, a signal descriptor and an instruction descriptor are acquired at each sampling point; according to the signal descriptor and the instruction descriptor, the signal descriptor and the instruction descriptor are combined to obtain a signal instruction integrated descriptor; and finally, one or more target signal sampling points are identified in the monitoring signal segment based on the signal instruction integrated descriptor and the signal descriptor. The application can acquire the correlation of a signal descriptor and an instruction descriptor after the signal descriptor and the instruction descriptor are extracted, combine the signal descriptor and the instruction descriptor based on the correlation of the descriptors, complete high-precision integration of the signal descriptor and the instruction descriptor, increase the target signal identification precision, and realize high-precision visualization.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of data processing, in particular to a monitoring signal visualization method applied to an implantable medical monitoring sensor. BACKGROUND

[0002] With the continuous progress of medical technology, implantable medical monitoring sensors are increasingly widely used in modern medical fields. These sensors can acquire various physiological signals inside the human body in real time, such as electrocardiogram signals, blood glucose levels, neural activity signals, etc., providing key data support for disease diagnosis, treatment, and health condition monitoring. In the process of medical monitoring, the physiological signal data monitored is large in quantity and complex. Traditional monitoring signal processing methods often focus on simple filtering, amplification, and basic feature extraction of the signal, and lack effective means for how to effectively utilize the instruction information related to the monitoring purpose to accurately process the monitoring signal. For example, when monitoring heart activity, although electrocardiogram signals can be collected, how to combine requirements such as the doctor's focus on a specific heart condition focus area (such as focusing on the changes of electrocardiogram waveform in a specific time period, special attention to a certain frequency band signal, etc.) and organically integrate these requirements with the actually collected signals to improve the accuracy of identifying abnormal heart activity signals is still a challenge. SUMMARY

[0003] The purpose of the present application is to provide a monitoring signal visualization method applied to an implantable medical monitoring sensor. The present application is implemented in the following way:

[0004] In a first aspect, the present application provides a monitoring signal visualization method applied to an implantable medical monitoring sensor, comprising:

[0005] acquiring a target monitoring signal data stream and a detection guide instruction corresponding to the target monitoring signal data stream, the target monitoring signal data stream comprising one or more monitoring signal segments;

[0006] performing implicit representation on the monitoring signal segments to obtain signal descriptors of one or more feature axes, and performing implicit representation on the detection guide instruction to obtain an instruction descriptor;

[0007] acquiring a descriptor correlation coefficient of the signal descriptor and the instruction descriptor at each sampling point, the descriptor correlation coefficient representing the correlation degree of the signal descriptor and the instruction descriptor at the same sampling point;

[0008] merging the signal descriptor and the instruction descriptor according to the descriptor correlation coefficient to obtain a signal-instruction integrated descriptor;

[0009] Based on the signal instruction integrated descriptor and the signal descriptor, one or more target signal sampling points are identified in the monitoring signal segment;

[0010] Based on the identified one or more target signal sampling points, the target monitoring signal data stream is displayed according to a preset visual element, and a display monitoring signal is obtained.

[0011] Optionally, the descriptor correlation coefficient of the signal descriptor and the instruction descriptor at each sampling point includes:

[0012] The signal descriptor and the instruction descriptor are projected into the same feature domain to obtain a target signal descriptor of each feature axis and a target instruction descriptor corresponding to the target signal descriptor;

[0013] The sampling point signal descriptor corresponding to each sampling point is extracted from the target signal descriptor, and the descriptor correlation coefficient between the sampling point signal descriptor and the target instruction descriptor is obtained.

[0014] Optionally, the instruction descriptor includes one or more instruction unit descriptors, and the projection of the signal descriptor and the instruction descriptor into the same feature domain to obtain a target signal descriptor of each feature axis and a target instruction descriptor corresponding to the target signal descriptor includes:

[0015] The number of descriptor components is determined in the signal descriptor to obtain the number of signal components, and the number of descriptor components is determined in the instruction descriptor to obtain the number of instruction components;

[0016] According to the number of signal components and the number of instruction components, the target component number of the feature domain corresponding to each feature axis is determined, and the instruction unit descriptors are merged to obtain a merged instruction descriptor;

[0017] The number of components of the signal descriptor and the number of components of the merged instruction descriptor are respectively transformed into the target component number to obtain a target signal descriptor of each feature axis and a target instruction descriptor corresponding to the target signal descriptor;

[0018] The descriptor correlation coefficient between the sampling point signal descriptor and the target instruction descriptor includes:

[0019] The descriptor under each descriptor component is extracted from the sampling point signal descriptor to obtain a signal component descriptor;

[0020] The descriptor under the descriptor component corresponding to the signal component descriptor is selected from the target instruction descriptor to obtain an instruction component descriptor;

[0021] determine a descriptor correlation coefficient between the sample point signal descriptor and the target instruction descriptor according to the signal component descriptor and the instruction component descriptor.

[0022] Optionally, the determining the descriptor correlation coefficient between the sample point signal descriptor and the target instruction descriptor according to the signal component descriptor and the instruction component descriptor comprises:

[0023] merging the signal component descriptor and the instruction component descriptor to obtain a merged component descriptor corresponding to each descriptor component;

[0024] fusing the merged component descriptor of each sample point to obtain a target cumulative descriptor corresponding to each sample point;

[0025] merging the sample point signal descriptor and the target instruction descriptor to obtain a contrast cumulative descriptor, and obtaining a proportion between the contrast cumulative descriptor and the target cumulative descriptor to obtain a descriptor correlation coefficient corresponding to each sample point;

[0026] the merging the signal descriptor and the instruction descriptor according to the descriptor correlation coefficient to obtain a signal instruction integrated descriptor comprises:

[0027] weighting the sample point signal descriptor according to the descriptor correlation coefficient to obtain a basic signal instruction integrated descriptor corresponding to each sample point;

[0028] merging the basic signal instruction integrated descriptors of the same feature axis to obtain a signal instruction integrated descriptor of each feature axis.

[0029] Optionally, the identifying one or more target signal sample points in the monitoring signal segment based on the signal instruction integrated descriptor and the signal descriptor comprises:

[0030] sequentially arranging the signal descriptors according to the feature axes of the signal descriptors, and selecting a signal descriptor of a target feature axis from the signal descriptors according to the sequentially arranged result to obtain a current signal descriptor;

[0031] selecting a signal instruction integrated descriptor corresponding to the target feature axis from the signal instruction integrated descriptors to obtain a current signal instruction integrated descriptor;

[0032] extracting a target signal descriptor from the monitoring signal segment based on the current signal instruction integrated descriptor and the current signal descriptor, and determining one or more target signal sample points in the monitoring signal segment according to the target signal descriptor;

[0033] wherein the extracting the target signal descriptor from the monitoring signal segment based on the current signal instruction integrated descriptor and the current signal descriptor comprises:

[0034] optimizing the preset target signal descriptor according to the current signal instruction integrated descriptor and the current signal descriptor, and determining the optimized target signal descriptor as the preset target signal descriptor;

[0035] cycling to the step of selecting the target feature axis from the signal descriptor according to the sequentially arranged result until each signal descriptor is the current signal descriptor, and obtaining the target signal descriptor.

[0036] Optionally, the optimizing the preset target signal descriptor according to the current signal instruction integrated descriptor and the current signal descriptor comprises:

[0037] optimizing the preset target signal descriptor according to the current signal instruction integrated descriptor and the current signal descriptor, and obtaining a basic target signal descriptor, and determining the basic target signal descriptor as the preset target signal descriptor;

[0038] cycling to the step of optimizing the preset target signal descriptor according to the current signal instruction integrated descriptor and the current signal descriptor until a preset optimization number is met, and obtaining an optimized target signal descriptor.

[0039] Optionally, the preset target signal descriptor comprises one or more sub-target signal descriptors, and the optimizing the preset target signal descriptor according to the current signal instruction integrated descriptor and the current signal descriptor to obtain a basic target signal descriptor comprises:

[0040] performing adaptive weight allocation on the sub-target signal descriptor to obtain a candidate target signal descriptor;

[0041] performing cross-modal weight allocation on the candidate target signal descriptor, the current signal instruction integrated descriptor and the current signal descriptor to obtain a current target signal descriptor;

[0042] projecting the current target signal descriptor to a preset target representation domain to obtain the basic target signal descriptor.

[0043] Optionally, the implicitly representing the monitoring signal segment to obtain one or more feature axis signal descriptors comprises:

[0044] performing multi-level implicit representation on the monitoring signal segment to obtain one or more feature level basic signal descriptors;

[0045] normalizing the basic signal descriptor to obtain a candidate signal descriptor of one or more feature axes;

[0046] fusing the candidate signal descriptor based on a preset feature fusion path to obtain a signal descriptor of one or more feature axes.

[0047] Optionally, the normalizing the basic signal descriptor to obtain a candidate signal descriptor of one or more feature axes comprises:

[0048] selecting a target basic signal descriptor from the basic signal descriptor according to the feature hierarchy, and adjusting a dimension of the target basic signal descriptor to obtain an adjusted signal descriptor;

[0049] combining the adjusted signal descriptor with a fixed position code to obtain a combined signal descriptor, the combined signal descriptor comprising one or more sample point signal descriptors;

[0050] performing adaptive weight distribution on the sample point signal descriptor to obtain a weighted signal descriptor, and optimizing the basic signal descriptor based on the weighted signal descriptor to obtain a candidate signal descriptor of one or more feature axes;

[0051] The preset feature fusion path comprises a first feature fusion path and a second feature fusion path, and the fusing the candidate signal descriptor based on a preset feature fusion path to obtain a signal descriptor of one or more feature axes comprises:

[0052] sequentially arranging the candidate signal descriptors according to a descriptor dimension of the candidate signal descriptors to obtain a first dimension sequential arrangement result;

[0053] fusing the candidate signal descriptors based on the first dimension sequential arrangement result according to the first feature fusion path to obtain one or more candidate fusion descriptors;

[0054] sequentially arranging the candidate fusion descriptors according to a descriptor dimension of the candidate fusion descriptors to obtain a second dimension sequential arrangement result;

[0055] fusing the candidate fusion descriptors based on the second dimension sequential arrangement result according to the second feature fusion path to obtain one or more fusion descriptors, and determining each fusion descriptor as a signal descriptor of a feature axis;

[0056] The first dimension sequence arrangement result is used to select one or more candidate signal descriptors from the candidate signal descriptors, and the candidate signal descriptors are fused according to the first feature fusion path to obtain one or more candidate fusion descriptors, including:

[0057] The candidate signal descriptor with the smallest dimension is selected from the candidate signal descriptors to obtain a current candidate signal descriptor, and the current candidate signal descriptor is determined as a first candidate fusion descriptor;

[0058] The next candidate signal descriptor of the current candidate signal descriptor is selected from the candidate signal descriptors according to the first dimension sequence arrangement result to obtain a first tentative fusion descriptor of the current candidate signal descriptor;

[0059] The current candidate signal descriptor and the first tentative fusion descriptor are merged to obtain a second candidate fusion descriptor, and the first tentative fusion descriptor is determined as the current candidate signal descriptor;

[0060] The step of selecting the next candidate signal descriptor of the current candidate signal descriptor from the candidate signal descriptors according to the first dimension sequence arrangement result is looped until the candidate signal descriptors are fused to obtain one or more candidate fusion descriptors, and the candidate fusion descriptors include the first candidate fusion descriptor and the second candidate fusion descriptor.

[0061] In a second aspect, the present application provides a computer system, including: one or more processors; a memory; one or more computer programs; wherein the one or more computer programs are stored in the memory and configured to be executed by the one or more processors, and when the one or more computer programs are executed by the processors, the method described above is implemented.

[0062] The beneficial effects of the present application: the present application obtains the target monitoring signal data stream and the detection guide instruction corresponding to the target monitoring signal data stream, and then implicitly represents the monitoring signal segment in the target monitoring signal data stream, obtains the signal descriptor of one or more feature axes, and implicitly represents the detection guide instruction to obtain the instruction descriptor, then obtains the descriptor correlation coefficient of the signal descriptor and the instruction descriptor at each sampling point, merges the signal descriptor and the instruction descriptor according to the descriptor correlation coefficient, obtains the signal instruction integrated descriptor, and then identifies one or more target signal sampling points in the monitoring signal segment based on the signal instruction integrated descriptor and the signal descriptor. Because the present application can extract the signal descriptor and the instruction descriptor, obtain the descriptor correlation coefficient of the signal descriptor and the instruction descriptor at each sampling point, and merge the signal descriptor and the instruction descriptor based on the descriptor correlation coefficient, the high-precision integration between the signal descriptor and the instruction descriptor is completed through the sampling point granularity merging strategy during the merging, the precision of target signal identification is increased, and high-precision visualization is realized. BRIEF DESCRIPTION OF DRAWINGS

[0063] Figure 1 is a flowchart of a monitoring signal visualization method applied to an implantable medical monitoring sensor provided by an embodiment of the present application.

[0064] Figure 2 is a component schematic diagram of a computer system provided by an embodiment of the present application. DETAILED DESCRIPTION

[0065] The execution subject of the monitoring signal visualization method applied to the implantable medical monitoring sensor in the embodiment of the present application is a computer system, including but not limited to a server, a personal computer, a notebook computer, a tablet computer, a smart phone and the like. As shown in Figure 1 The method comprises the following steps:

[0066] Step S100: obtaining a target monitoring signal data stream and a detection guide instruction corresponding to the target monitoring signal data stream, the target monitoring signal data stream comprising one or more monitoring signal segments.

[0067] In the embodiment of the present application, the target monitoring signal data stream is the core data source of the entire visualization process, which contains one or more monitoring signal segments. The target monitoring signal data stream herein can be understood as a series of continuous signal data collected from an implantable medical monitoring sensor. For example, in the application scenario of a cardiac implantable monitoring sensor, the target monitoring signal data stream can be continuous signal data about cardiac electrical activity. These signal data are not single and unstructured, but are divided into multiple monitoring signal segments. The monitoring signal segment is the partial signal data obtained by dividing the continuous signal data according to certain rules or time intervals. For example, the cardiac electrical activity signal can be divided according to a time interval of every 5 seconds, and each 5-second signal data is a monitoring signal segment. The detection guide instruction is instruction information corresponding to the target monitoring signal data stream. It plays a guiding and assisting role in the entire visualization process. The detection guide instruction can contain multiple information, such as parameter information of data acquisition (such as acquisition frequency, acquisition accuracy, etc.), some pre-analysis results of the target monitoring signal, or instructions related to the monitoring purpose, etc. Continuing to take the cardiac implantable monitoring sensor as an example, the detection guide instruction can contain information such as the current focus area of cardiac electrical activity monitoring (such as a specific electrocardiogram waveform interval), the specific acquisition frequency set when acquiring the cardiac electrical activity signal (such as acquiring a signal once every millisecond), etc.

[0068] The computer system can obtain the target monitoring signal data stream and the detection guide instruction through various technical means. For the acquisition of the target monitoring signal data stream, the computer system can receive data through the communication interface established with the implantable medical monitoring sensor. This communication interface can be wired, such as connecting the sensor and the computer system through a dedicated medical data transmission line; or wireless, such as using Bluetooth, Wi-Fi, etc. wireless communication technology, the sensor sends the collected signal data to the computer system. For the acquisition of the detection guide instruction, on the one hand, the relevant instructions can be input into the computer system by the operator during the initialization or configuration process of the sensor; on the other hand, if the detection guide instruction is the pre-analysis result generated by the internal algorithm of the sensor, etc. information, the computer system can directly read these instruction information from the sensor or the storage device related thereto.

[0069] Step S200: implicitly representing the monitoring signal segment to obtain one or more feature axis signal descriptors, and implicitly representing the detection guide instruction to obtain an instruction descriptor.

[0070] In the embodiments of the present application, step S200 performs implicit representation on the monitoring signal segment to obtain a signal descriptor of one or more feature axes, and performs implicit representation on the detection guide instruction to obtain an instruction descriptor. For the implicit representation of the monitoring signal segment, it is a process of converting the original monitoring signal segment into a more characteristic representation form. Taking an implantable blood glucose monitoring sensor as an example, the original monitoring signal segment may be electrical signal data reflecting the change of blood glucose level over time, which is difficult to interpret directly and contains a large amount of complex information. The computer system performs implicit representation on it through a specific algorithm, aiming to extract information that can better describe the characteristics of the signal, and finally obtain a signal descriptor of one or more feature axes. The feature axis can be understood as a benchmark for describing the characteristics of the signal from different dimensions or angles. For example, in describing the blood glucose monitoring signal, one feature axis may be related to the rate of change of blood glucose concentration, and another feature axis may be related to the fluctuation amplitude of blood glucose concentration. These different feature axes combined together can more comprehensively describe the characteristics of the blood glucose monitoring signal, and the description corresponding to each feature axis is the signal descriptor.

[0071] In terms of implicit representation of the detection guide instruction to obtain the instruction descriptor, the detection guide instruction contains various information related to monitoring. Assuming that in the scenario of implantable blood pressure monitoring, the detection guide instruction contains information such as the time interval of blood pressure monitoring and the normal range of blood pressure. The computer system needs to convert these instructions into an implicit representation form to obtain the instruction descriptor. This instruction descriptor can exist in a form corresponding or associated with the signal descriptor, so as to perform subsequent correlation analysis and other operations.

[0072] In order to realize the implicit representation of the monitoring signal segment, the computer system can adopt various technical means. A common technical means is to use feature extraction algorithms in signal processing. For example, for a monitoring signal segment containing time series information, a wavelet transform algorithm can be used. Wavelet transform can decompose the signal into components of different scales and frequencies, and by analyzing these components, information related to different feature axes can be determined, and then the signal descriptor can be constructed. Another technical means is a feature learning method based on machine learning. The computer system can use an autoencoder in a deep learning model, take the monitoring signal segment as input, and map it to a low-dimensional feature space through the encoding layer of the autoencoder. The vector in this low-dimensional space can be used as the signal descriptor.

[0073] For detecting the implicit representation of the guidance instruction, the computer system can adopt a rule-based conversion method. According to different types of information in the guidance instruction, corresponding conversion rules are formulated. For example, for numerical instruction information (such as the time interval of blood pressure monitoring), a certain mathematical formula can be used for conversion; for interval instruction information (such as the normal range of blood pressure), it can be converted into an interval representation form matching the signal characteristics, so as to obtain the instruction descriptor.

[0074] Step S300: Obtain the descriptor correlation coefficient of the signal descriptor and the instruction descriptor at each sampling point, which represents the correlation degree of the signal descriptor and the instruction descriptor at the same sampling point.

[0075] In the embodiment of the present application, the descriptor correlation coefficient of the signal descriptor and the instruction descriptor at each sampling point is obtained in step S300, which represents the correlation degree of the signal descriptor and the instruction descriptor at the same sampling point.

[0076] In the medical monitoring scenario, taking the implantable heart monitoring sensor as an example, the signal descriptor is the description information about the characteristics of heart activity obtained by implicitly representing the heart monitoring signal segment, which describes the heart signal from different feature axes. The instruction descriptor is the result of implicitly representing the detection guidance instruction, which may contain information such as the key attention period of heart monitoring, the expected occurrence of specific waveform characteristics, etc. At each sampling point, there is a certain correlation degree between the two, and the descriptor correlation coefficient is an index for quantifying the correlation degree.

[0077] For example, at a certain sampling point, if the signal descriptor shows that an abnormal peak value of cardiac electrical activity occurs, and the instruction descriptor mentions that there is a requirement for key monitoring of the waveform characteristics related to the abnormal peak value, then the correlation degree of the two at this sampling point may be high, and the corresponding descriptor correlation coefficient will also be large. Conversely, if the characteristics in the signal descriptor and the requirements in the instruction descriptor are quite different, the correlation coefficient will be small. In order to obtain the descriptor correlation coefficient, the computer system can use a variety of technical means. One feasible method is based on the vector space model. The computer system regards the signal descriptor and the instruction descriptor as vectors in the vector space, and the dimension of each vector corresponds to different features or information components. Then the similarity between the vectors is calculated to determine the descriptor correlation coefficient. For example, the cosine similarity algorithm can be used to calculate the cosine value of the angle between the signal descriptor vector and the instruction descriptor vector, which can be used as the descriptor correlation coefficient. If the cosine value is close to 1, it means that the directions of the two vectors are similar, i.e. the correlation degree is high; if the cosine value is close to 0, it means that the correlation degree is low.

[0078] Another technical means is the method based on probability statistics. The computer system can analyze the probability of the signal descriptor and the instruction descriptor appearing with a certain feature combination in a large amount of historical monitoring data. For the current sampling point, the descriptor correlation coefficient is determined according to the probability distribution of similar situations in the historical data. For example, if in the historical data, when the instruction descriptor contains a certain specific monitoring requirement, the signal descriptor appears with a high probability of matching features, then if a similar situation also appears at the current sampling point, a higher descriptor correlation coefficient will be given.

[0079] Step S400: merging the signal descriptor and the instruction descriptor according to the descriptor correlation coefficient to obtain a signal-instruction integrated descriptor.

[0080] In the embodiment of the present application, step S400 merges the signal descriptor and the instruction descriptor according to the descriptor correlation coefficient to obtain a signal-instruction integrated descriptor.

[0081] Taking an implantable neural activity monitoring sensor as an example, the signal descriptor is the description information about the characteristics of the neural signal obtained after the neural activity monitoring signal segment is implicitly represented, which reflects the performance of the neural activity on each feature axis, such as the quantitative description of the characteristics of the neural impulse, such as the frequency, amplitude, etc. The instruction descriptor is the result of the implicit representation of the detection guide instruction, which may contain information such as the attention to a specific neural activity mode and the detection requirement of a specific frequency band neural signal. The descriptor correlation coefficient characterizes the correlation degree of the signal descriptor and the instruction descriptor at each sampling point.

[0082] The computer system combines the signal descriptor and the instruction descriptor according to the descriptor correlation coefficient. It is assumed that the descriptor correlation coefficient is high at a certain sampling point, which indicates that the signal descriptor and the instruction descriptor at the point have strong correlation. For example, the instruction descriptor requires focusing on the neural activity of a certain frequency band, and the signal descriptor at the sampling point shows that the neural activity of the frequency band is active. At this time, the computer system will perform a more close combination operation on the signal descriptor and the instruction descriptor at the sampling point according to the high correlation coefficient. To achieve this combination, the computer system can use the weighted summation technique. The signal descriptor and the instruction descriptor are regarded as vectors, and the weights are determined according to the descriptor correlation coefficient. For example, for a sampling point with a correlation coefficient of 0.8, the signal descriptor is assigned a weight of 0.8 and the instruction descriptor is assigned a weight of 0.2. Then, the weighted signal descriptor and the weighted instruction descriptor are added to obtain the combined result as the part of the signal-instruction integrated descriptor at the sampling point. Another technical means is based on a feature fusion algorithm. If the signal descriptor and the instruction descriptor are regarded as objects with different feature sets, the computer system can determine the fusion proportion and manner according to the descriptor correlation coefficient. For example, in the case of high correlation coefficient, more information of common features in the signal descriptor and the instruction descriptor is retained, and these features are combined in an optimized manner to construct the signal-instruction integrated descriptor.

[0083] Through the combination operation, the computer system can organically combine the signal descriptor and the instruction descriptor to form the signal-instruction integrated descriptor. The integrated descriptor contains the feature information of the monitoring signal itself and the related requirements and expectations of the detection guide instruction, which provides a more comprehensive and accurate basis for subsequent accurate identification of the target signal sampling point in the monitoring signal segment, and helps to realize high-precision visualization of the monitoring signal.

[0084] Step S500: identifying one or more target signal sampling points in the monitoring signal segment based on the signal-instruction integrated descriptor and the signal descriptor.

[0085] In the embodiment of the present application, step S500 identifies one or more target signal sampling points in the monitoring signal segment based on the signal-instruction integrated descriptor and the signal descriptor.

[0086] For example, in the case of an implantable heart monitoring sensor, the signal descriptor contains information describing the heart monitoring signal segment from different feature axes, such as the frequency variation feature of heartbeats, the morphology feature of electrocardiogram waveforms, and so on. The signal instruction integrated descriptor is the result of integrating the signal descriptor and the instruction descriptor, which not only contains the feature information of the signal itself, but also incorporates relevant requirements in the detection guide instructions, such as the focus on abnormal heart waveform patterns or the expected state of heart activity within a specific period. The computer system identifies the target signal sampling points by analyzing these two descriptors. The target signal sampling points are sampling points that have special significance or meet specific requirements in the entire monitoring signal segment. For example, in heart monitoring, if the signal instruction integrated descriptor contains a requirement to focus on a certain abnormal electrocardiogram waveform (such as a premature beat waveform), and the signal descriptor presents similar features at certain sampling points, these sampling points may be identified as target signal sampling points.

[0087] To achieve this identification, the computer system can use pattern matching techniques. The known target signal pattern (corresponding to the pattern of the relevant requirements extracted from the signal instruction integrated descriptor) is matched with the signal descriptor in the monitoring signal segment. For example, for electrocardiogram waveform pattern matching, the characteristic points of the waveform (such as the positions and amplitudes of wave peaks and troughs) can be used for comparison. If the waveform characteristic points in the signal descriptor match the characteristic points in the target signal pattern within a certain error range at a certain sampling point, that sampling point may be identified as a target signal sampling point. Another technical means is based on threshold judgment. The computer system sets a threshold according to the requirements in the signal instruction integrated descriptor. For example, if the focus is on the case where the heart rate is too high, a frequency threshold is set according to the normal heart rate range. When the heart rate in the signal descriptor exceeds this threshold at a certain sampling point, that sampling point may be identified as a target signal sampling point.

[0088] The computer system accurately identifies the target signal sampling points through step S500, which lays the foundation for subsequent display processing of the target monitoring signal data stream based on the pre-set visualization elements, and helps to more accurately present valuable monitoring information in a visualized manner, meeting the needs of implantable medical monitoring sensor monitoring signal visualization.

[0089] Step S600: Based on the one or more target signal sampling points identified, the target monitoring signal data stream is displayed and processed according to the pre-set visualization elements, and the displayed monitoring signal is obtained.

[0090] In the embodiment of the present application, step S600 displays and processes the target monitoring signal data stream according to the preset visual elements based on the identified one or more target signal sampling points, thereby obtaining the displayed monitoring signal. Taking the implantable blood glucose monitoring sensor as an example, the target monitoring signal data stream contains a series of monitoring data about the change of blood glucose level, and these data are divided into multiple monitoring signal segments. In the previous steps, the computer system has identified the target signal sampling points, which have special significance in the entire blood glucose monitoring data, such as possibly corresponding to the rapid change points or abnormal value points of blood glucose concentration.

[0091] The preset visual elements are basic components for constructing the visual representation, including but not limited to color, shape, line style, icon, etc. For example, when representing the blood glucose monitoring signal, different colors can be set to represent different blood glucose concentration ranges. The signal part corresponding to the normal blood glucose concentration range is represented by green, the signal part corresponding to the high blood glucose concentration range is represented by red, and the signal part corresponding to the low blood glucose concentration range is represented by yellow. In terms of shape, a circular icon can be used to mark the target signal sampling points, which may be the time when the blood glucose concentration mutates or exceeds the normal range. The line style can also be used to distinguish different types of blood glucose change trends, such as using a solid line to represent an upward trend and a dashed line to represent a downward trend.

[0092] The computer system displays the target monitoring signal data stream according to the preset visualization elements. First, for the entire target monitoring signal data stream, the computer system arranges the data in chronological order or other logical order. Then, according to the blood glucose concentration corresponding to each data point, it gives it the corresponding color. For data points in the normal blood glucose concentration range, set their color to green; for data points in the high blood glucose concentration range, set their color to red; for data points in the low blood glucose concentration range, set their color to yellow. For the identified target signal sampling points, the computer system superimposes the corresponding icons (such as circular icons) onto the signal graph. In this way, when medical personnel or patients view the visualization results, they can see at a glance which moments are the moments when the blood glucose concentration changes abnormally. At the same time, according to the line style, the computer system draws a line representing the trend of blood glucose concentration. If the blood glucose concentration is rising in a certain time period, connect the data points in that time period with a solid line; if it is a downward trend, connect it with a dashed line. In order to achieve this display processing, the computer system can use a variety of technical means. In terms of graph drawing, you can use a graphics library, such as the Matplotlib library in Python. The Matplotlib library provides a wealth of functions and tools for creating various types of graphs, including line graphs, scatter plots, etc. The computer system can use the functions of this library to create a line graph representing the blood glucose monitoring signal, set the color, style of the line, and the markers of the data points, etc.

[0093] In processing the mapping relationship between the data and the visualization elements, data processing and conversion algorithms can be used. For example, the blood glucose concentration values are converted into corresponding color values through a mapping function. This mapping function can be linear or customized according to actual needs. For the superposition of the target signal sampling points and the icons, coordinate positioning algorithms can be used. The computer system calculates the coordinate position of the target signal sampling points in the graph according to their positions in the data sequence, and then draws the icons on the corresponding coordinate positions. In addition, in order to improve the interactivity of the visualization, the computer system can also use interactive visualization technology. For example, when the user hovers the mouse over a certain data point or target signal sampling point, detailed blood glucose concentration values, collection times, and other information are displayed. This can be achieved through front-end technologies such as JavaScript on the Web interface, or by using corresponding interactive components in the local application. Through step S600, the computer system converts the originally complex and difficult-to-understand target monitoring signal data stream of the implantable medical monitoring sensor into an intuitive and easy-to-understand display monitoring signal. This visualization representation helps medical personnel quickly and accurately analyze the monitoring results, timely detect abnormal conditions and make corresponding medical decisions, and also facilitates patients to understand their own physical conditions. For different types of implantable medical monitoring sensors, such as implantable heart monitoring sensors, blood pressure monitoring sensors, etc., similar methods can be used to set appropriate visualization elements and perform display processing according to the monitoring data characteristics and clinical needs of each type, thereby achieving efficient medical monitoring signal visualization.

[0094] As an implementation manner, step S300 of obtaining the descriptor correlation between the signal descriptor and the instruction descriptor at each sampling point can include:

[0095] Step S310: projecting the signal descriptor and the instruction descriptor into the same representation domain to obtain the target signal descriptor of each feature axis and the target instruction descriptor corresponding to the target signal descriptor;

[0096] Step S320: extracting the sampling point signal descriptor corresponding to each sampling point from the target signal descriptor, and obtaining the descriptor correlation between the sampling point signal descriptor and the target instruction descriptor.

[0097] In step S300 of the embodiment of the present application, the specific implementation manner includes steps S310 and S320, which are mainly the operations performed by the computer system to obtain the descriptor correlation between the signal descriptor and the instruction descriptor at each sampling point.

[0098] In step S310, the computer system projects the signal descriptor and the instruction descriptor into the same representation domain, so as to obtain the target signal descriptor and the target instruction descriptor corresponding to each feature axis. This process is to compare and correlate the signal descriptor and the instruction descriptor under the same standard or framework.

[0099] Taking an implantable electrocardiogram monitoring sensor as an example, the signal descriptor can include descriptions of electrocardiogram signals in different features, such as the amplitude feature of QRS complex and the morphology feature of ST segment. These features describe the electrocardiogram signal from different angles and are in different original representation forms. The instruction descriptor can include information about the key attention area of the electrocardiogram signal (such as the monitoring requirement of electrocardiogram activity in a specific time period) or abnormal waveform feature (such as the attention requirement for the ST segment change that can indicate myocardial ischemia), which also has its original representation form.

[0100] The computer system projects them into the same representation domain. Assuming that this representation domain is a three-dimensional space based on frequency-amplitude-time. For the QRS complex amplitude feature in the signal descriptor, the computer system will correspond its amplitude information to the amplitude dimension in this three-dimensional space according to certain conversion rules. For the ST segment morphology feature, it can be linked to the frequency and time dimensions through certain mathematical transformation. For the key attention time period requirement in the instruction descriptor, it is also converted to the time dimension in this three-dimensional space, and the attention requirement for the abnormal waveform feature is converted to the frequency-amplitude dimension through related algorithms. In this way, in this three-dimensional representation domain, the computer system obtains the target signal descriptor and the target instruction descriptor corresponding to each feature axis (here, the features represented by each coordinate axis in the three-dimensional space).

[0101] For numerical features in the signal descriptor and the instruction descriptor, coordinate transformation algorithms can be used to convert them to the representation domain under the same coordinate system. For example, the original electrocardiogram amplitude feature is scaled according to a certain proportion and translated along the coordinate axis, so as to adapt to the amplitude coordinate axis in the new representation domain. For some features with complex logical relationships, a feature mapping function is constructed. For example, for the attention requirement for a specific electrocardiogram waveform combination in the instruction descriptor, a mapping function is established to convert this logical requirement into a form that can be compared with the signal descriptor in the same representation domain. This mapping function can be based on the analysis and statistical rules of a large amount of historical electrocardiogram data.

[0102] In step S320, after projecting the target signal descriptor and the target instruction descriptor, the computer system extracts a sampling point signal descriptor corresponding to each sampling point from the target signal descriptor, and obtains a descriptor correlation coefficient between the sampling point signal descriptor and the target instruction descriptor. The purpose of this step is to quantify the correlation degree between the signal descriptor and the instruction descriptor at each sampling point. Continuing with the above example of ECG monitoring, after the signal descriptor and the instruction descriptor have been projected into the same representation domain, each sampling point in the target signal descriptor contains feature information of the ECG signal in this representation domain. For example, at a certain sampling point, the target signal descriptor may represent the specific numerical value of the ECG signal in the frequency-amplitude-time three-dimensional representation domain at that time, such as the amplitude value at a specific frequency and the corresponding time point information. The computer system needs to obtain the correlation coefficient between this sampling point signal descriptor and the target instruction descriptor. Assuming that the target instruction descriptor focuses on the depression of the ST segment (this requirement has been converted into the same representation domain), if the ST segment-related features in the target signal descriptor show a certain degree of depression at this sampling point, and the degree of coincidence with the attention requirements for ST segment depression in the target instruction descriptor (such as the amplitude range, duration, etc. requirements for depression) is high, then the descriptor correlation coefficient of this sampling point will be larger; otherwise, if it does not coincide or differs greatly, the correlation coefficient will be smaller.

[0103] In implementation, the sampling point signal descriptor and the target instruction descriptor can be regarded as vectors, and a vector similarity calculation method can be used to obtain the correlation coefficient. For example, the cosine similarity algorithm can be used. Assuming that the sampling point signal descriptor vector is and the target instruction descriptor vector is the cosine similarity is and this can be used as the descriptor correlation coefficient. This method is suitable for cases where the features in the representation domain can be represented in vector form.

[0104] Specifically, the correlation coefficient can be calculated according to pre-set rules. For example, for the requirement of the number of occurrences of certain specific waveforms in ECG monitoring, if the target instruction descriptor specifies the range of the number of occurrences of a certain waveform within a certain time window, the computer system counts the actual number of occurrences of the waveform in the sampling point signal descriptor within this time window, and then determines a correlation coefficient according to the matching degree of the actual number of occurrences with the specified range. For example, if the actual number of occurrences is completely within the specified range, the correlation coefficient is 1; if it exceeds the specified range by a certain percentage, the correlation coefficient is correspondingly reduced.

[0105] As an implementation, the instruction description sub includes one or more instruction unit description sub, the step S310 projects the signal description sub and the instruction description sub into the same representation domain, obtains the target signal description sub of each feature axis and the target instruction description sub corresponding to the target signal description sub, which can include:

[0106] The step S311: determines the number of description sub components in the signal description sub, obtains the number of signal components, and determines the number of description sub components in the instruction description sub, obtains the number of instruction components;

[0107] The step S312: according to the number of signal components and the number of instruction components, determines the target number of components of the representation domain corresponding to each feature axis, and merges the instruction unit description sub to obtain a merged instruction description sub;

[0108] The step S313: transforms the number of components of the signal description sub and the number of components of the merged instruction description sub into the target number of components respectively, obtains the target signal description sub of each feature axis and the target instruction description sub corresponding to the target signal description sub.

[0109] In the embodiment of the application, the operation of projecting the signal description sub and the instruction description sub into the same representation domain to obtain the target signal description sub and the target instruction description sub of each feature axis in the specific implementation of the step S310.

[0110] In the step S311, the computer system determines the number of description sub components in the signal description sub, thereby obtaining the number of signal components, and determines the number of description sub components in the instruction description sub, thereby obtaining the number of instruction components. Taking the implanted brain electrical monitoring sensor as an example, the signal description sub can contain data of multiple channels, and these channels collect brain electrical signals from different brain regions, and the data of each channel is a description sub component. Assuming that the brain electrical monitoring sensor has 8 channels for collecting signals of different brain regions, the number of signal components is 8. As for the instruction description sub, it can contain description sub components such as attention requirements for different frequency band brain electrical signals, monitoring requirements for specific brain region activities, etc. For example, the attention requirement for the alpha frequency band brain electrical signal in the instruction description sub, the monitoring requirement for the frontal lobe brain region activity, etc. can be regarded as different description sub components, and assuming that there are 3 such instruction description sub components, the number of instruction components is 3.

[0111] Then, in step S312, the computer system determines the target component number of the representation domain corresponding to each feature axis according to the signal component number and the instruction component number. Continuing with the example of the brain electrical monitoring, if the signal component number is 8 and the instruction component number is 3, the computer system can determine the target component number corresponding to each feature axis in the target representation domain according to certain optimization algorithm or pre-set rules. For example, considering the complexity of the signal and the instruction, the convenience of subsequent analysis and other factors, the computer system determines that the target component number corresponding to each feature axis is 5. At the same time, the computer system also needs to merge the instruction unit descriptors to obtain a merged instruction descriptor. Assuming that the instruction unit descriptors in the instruction descriptor include the requirement of paying attention to the alpha band brain electrical signal and the requirement of monitoring the activity of the frontal lobe brain region, the computer system merges them through a specific logical relationship. For example, if the activity of the alpha band brain electrical signal in the frontal lobe brain region has a specific correlation, the two instruction unit descriptors are merged according to this correlation to form a merged instruction descriptor.

[0112] Finally, in step S313, the computer system transforms the component number of the signal descriptor and the component number of the merged instruction descriptor into the target component number respectively, thereby obtaining the target signal descriptor of each feature axis and the target instruction descriptor corresponding to the target signal descriptor. Still taking the brain electrical monitoring as an example, for the signal descriptor, the original has 8 components, and the computer system transforms it into the target component number 5 by using data interpolation, feature extraction or dimension reduction techniques. For example, by using the dimension reduction technique of principal component analysis (PCA), the signal descriptor with 8 components is converted into the target signal descriptor with 5 components that retain the main information. For the merged instruction descriptor, the original has the component number after merging (assuming it is 1 complex merged instruction descriptor), which is also converted into the target component number 5 by using similar rules. For example, according to the pre-set mapping relationship, the information in the merged instruction descriptor can be distributed to the 5 target components in a certain proportion, thereby obtaining the target instruction descriptor.

[0113] In the determination of the component number, a statistical analysis method can be used. For example, a large amount of historical monitoring data is statistically analyzed to determine the reasonable signal component number and instruction component number under different monitoring scenarios. In terms of the merged instruction descriptor, a logical operation algorithm can be used. For instruction unit descriptors with logical association, such as "and", "or", "not" and the like, they are merged according to a specific logical relationship. In the transformation of the component number, as mentioned above, the principal component analysis (PCA) is used for the dimension reduction of the signal descriptor, and a linear interpolation algorithm can also be used to transform the component number of the signal descriptor. For the transformation of the instruction descriptor, a rule-based mapping algorithm can be used to convert the information of the instruction descriptor to the target instruction descriptor under the target component number according to the pre-set mapping rules.

[0114] As an implementation, the step S320 of obtaining the descriptor correlation coefficient between the sampling point signal descriptor and the target instruction descriptor can include:

[0115] The step S321 of extracting the descriptor under each descriptor component in the sampling point signal descriptor to obtain the signal component descriptor.

[0116] The step S322 of selecting the descriptor under the descriptor component corresponding to the signal component descriptor in the target instruction descriptor to obtain the instruction component descriptor.

[0117] The step S323 of determining the descriptor correlation coefficient between the sampling point signal descriptor and the target instruction descriptor according to the signal component descriptor and the instruction component descriptor.

[0118] In the step S321, the computer system extracts the descriptor under each descriptor component in the sampling point signal descriptor to obtain the signal component descriptor. The purpose of this step is to subdivide the sampling point signal descriptor according to its component structure, so as to perform more accurate comparison and correlation analysis with the target instruction descriptor in the subsequent steps. Taking the implantable heart monitoring sensor as an example, it is assumed that after the previous steps, the signal descriptor at a certain sampling point is a multi-component descriptor structure. For example, this sampling point signal descriptor contains information about different aspects of cardiac electrical activity, such as heart rate, amplitude of electrocardiogram waveform, frequency of electrocardiogram signal, etc. The computer system will perform extraction operation for each component in this sampling point signal descriptor. For the heart rate component, the computer system will extract the descriptor part related to the heart rate, which can include the value of the current heart rate, the trend of the heart rate change, etc.; for the electrocardiogram waveform amplitude component, the amplitude values of the main wave bands (such as QRS complex, T wave, etc.) of the electrocardiogram waveform at this sampling point will be extracted; for the electrocardiogram signal frequency component, the main frequency components of the electrocardiogram signal at this time will be extracted. These extracted descriptor parts from each component are the signal component descriptors.

[0119] Different data parsing algorithms can be designed for different types of signal description substructures. For example, if the signal description substructure is stored in a specific data format (e.g., JSON format), a computer system can write a parser to read the data under each component according to the format rules, thereby obtaining the signal component description substructure. For a signal description substructure represented in the form of a feature vector, a feature segmentation function can be constructed. For example, a vector containing multiple features (corresponding to different components) is segmented according to a pre-set feature index to segment the description substructure corresponding to each feature, thereby forming a signal component description substructure. In step S322, the computer system obtains the description substructure under the description substructure component corresponding to the signal component description substructure from the target instruction description substructure, thereby obtaining the instruction component description substructure. This step is to make the information obtained from the target instruction description substructure correspond to the signal component description substructure obtained in the previous step in terms of component structure, so as to perform correlation calculation.

[0120] Continuing with the example of heart monitoring, suppose that the target instruction description substructure contains multiple monitoring requirements for heart electrical activity, which are also organized according to different component structures. If the signal component description substructure obtained in step S321 contains information about heart rate, the computer system will search the target instruction description substructure for the monitoring requirement part related to heart rate. For example, the target instruction description substructure can contain a setting for the normal heart rate range, a requirement for the heart rate change rate limit, etc., which are instruction component description substructures for the heart rate component. Similarly, if the signal component description substructure contains information about the amplitude of the electrocardiogram waveform, the computer system will find the instruction component description substructure about the abnormal judgment standard for the amplitude of the electrocardiogram waveform (e.g., the normal range of QRS complex amplitude, the determination condition for T wave amplitude abnormality, etc.) in the target instruction description substructure. Specifically, an index matching algorithm can be used, such as establishing an index mechanism to set an index identifier for each component in the target instruction description substructure. The computer system finds the corresponding instruction component description substructure by index matching according to the component type of the signal component description substructure. For example, the index for the heart rate related instruction component is set to 1, and when the heart rate component is identified from the signal component description substructure, the computer system obtains the instruction component description substructure by searching the part with index 1. The semantic analysis technique is used to understand the semantic information in the target instruction description substructure, thereby determining the instruction component description substructure corresponding to the signal component description substructure. For example, for an instruction description substructure expressed in natural language (e.g., "focus on whether the heart rate is within the range of 60-100 times / minute"), the computer system identifies the part related to heart rate through semantic analysis, as the instruction component description substructure.

[0121] In step S323, the computer system determines the descriptor correlation coefficient between the signal descriptor and the target instruction descriptor according to the signal component descriptor and the instruction component descriptor. This step is to comprehensively analyze the signal component descriptor and the instruction component descriptor obtained above, so as to quantify the correlation degree between them.

[0122] Still taking the heart monitoring as an example, it is assumed that in a sampling point, the heart rate is 80 times per minute in the signal component descriptor, the heart rate change trend is stable, and the values of the amplitude of the ECG waveform in each main wave band are within the normal range; and the normal heart rate range is 60-100 times per minute in the target instruction descriptor, and the ECG waveform amplitude is required to be normal. In this case, the matching degree of the signal component descriptor and the instruction component descriptor is high in terms of the heart rate and the ECG waveform amplitude. The computer system determines the descriptor correlation coefficient according to the matching degree. If the matching degree is very high, a correlation coefficient close to 1 can be assigned. If there is some deviation, such as the heart rate is within the normal range but close to the boundary value, or the ECG waveform amplitude is normal but some wave bands are close to the abnormal range, the correlation coefficient can be about 0.8. If there is a significant mismatch, such as the heart rate is out of the normal range or the ECG waveform amplitude is significantly abnormal, the correlation coefficient can be very low, close to 0. For example, a rule-based scoring system can be implemented to determine the correlation coefficient. For example, different scoring rules are set for each component. For the heart rate component, if it is within the normal range, the distance from the center of the range is scored (the closer to the center, the higher the score). For the ECG waveform amplitude component, if all wave bands are normal, a full score is given, and if a wave band is close to the abnormal range, the score is deducted according to the closeness. Finally, the scores of each component are integrated to convert to the descriptor correlation coefficient through a mapping function. Alternatively, a probability statistical model can be used to calculate the correlation coefficient. The computer system analyzes a large amount of historical monitoring data to statistically calculate the correlation probability under different combinations of signal component descriptors and instruction component descriptors. For example, for different value range combinations of the heart rate and the ECG waveform amplitude, the probability of matching the instruction requirements in the past data is calculated, and then the corresponding probability value is looked up according to the signal component descriptor and the instruction component descriptor of the current sampling point as the descriptor correlation coefficient.

[0123] As an implementation, in step S323, the descriptor correlation coefficient between the signal descriptor and the target instruction descriptor is determined according to the signal component descriptor and the instruction component descriptor, which can include:

[0124] In step S3231, the signal component descriptor and the instruction component descriptor are merged to obtain a merged component descriptor corresponding to each descriptor component.

[0125] Step S3232: merging the merged component descriptor of each sampling point to obtain the target cumulative descriptor corresponding to each sampling point;

[0126] Step S3233: merging the sampling point signal descriptor and the target instruction descriptor to obtain the contrast cumulative descriptor, and obtaining the ratio between the contrast cumulative descriptor and the target cumulative descriptor to obtain the descriptor correlation coefficient corresponding to each sampling point.

[0127] In step S3231, the computer system merges the signal component descriptor and the instruction component descriptor to obtain the merged component descriptor corresponding to each descriptor component. Taking the implantable neural activity monitoring sensor as an example, the signal component descriptor can include the frequency, amplitude, and conduction velocity of the neural impulse, and the instruction component descriptor can include the expected range or specific requirements of these neural activity characteristics. For example, the neural impulse frequency in the signal component descriptor is 100 times per second, the amplitude is 5 millivolts, and the conduction velocity is 2 meters per second, while the expected range of the frequency in the instruction component descriptor is 80-120 times per second, the amplitude requirement is greater than 3 millivolts, and the conduction velocity requirement is between 1.5-2.5 meters per second. The computer system merges the signal component descriptor and the instruction component descriptor according to the component correspondence relationship. For the frequency component, the merged descriptor can include the actual frequency value 100 times per second and the expected range 80-120 times per second; for the amplitude component, the merged descriptor includes the actual amplitude 5 millivolts and the requirement greater than 3 millivolts; for the conduction velocity component, the merged descriptor includes the actual speed 2 meters per second and the requirement 1.5-2.5 meters per second. In this way, the merged component descriptor corresponding to each descriptor component is obtained.

[0128] To achieve this merging, the computer system can use the technical means of data structure combination. For example, a new data structure is created to combine the related data in the signal component descriptor and the instruction component descriptor according to the pre-defined order and format. For numerical data, the values of the two can be placed together to form a new structure; for range type data, the actual value and the expected range can be combined through a specific symbol or coding method.

[0129] Next, in step S3232, the computer system fuses the merged component descriptors of each sampling point to obtain the target cumulative descriptor corresponding to each sampling point. Continuing with the example of neural activity monitoring, assume that there are multiple merged component descriptors (e.g., the frequency, amplitude, and conduction velocity components described above) for a sampling point. The computer system fuses these merged component descriptors into a target cumulative descriptor using a fusion algorithm. For example, a weighted summation approach can be used for fusion, with different weights assigned to each component according to its importance in the overall neural activity description. Assume that the weight for frequency is 0.4, the weight for amplitude is 0.3, and the weight for conduction velocity is 0.3. For the merged component descriptor of the frequency component (100 Hz, 80-120 Hz), convert it into a numerical value (e.g., calculate a score based on the actual frequency and the expected range); similarly convert the amplitude and conduction velocity components, then multiply the converted values by their respective weights and sum them up to obtain the target cumulative descriptor corresponding to the sampling point.

[0130] In this process, the computer system can employ techniques including the design of the weighting algorithm. The weights of each component are determined based on the understanding and analysis of the monitoring target. Meanwhile, to convert the merged component descriptors into numerical values that can be used for weighting calculation, normalization techniques can also be employed. For example, the relationship between the actual value of frequency and the expected range is converted into a value between 0 and 1 through a normalization function, to facilitate subsequent weighting calculation.

[0131] Finally, in step S3233, the computer system merges the sampling point signal descriptor and the target instruction descriptor to obtain the control cumulative descriptor, and obtains the ratio between the control cumulative descriptor and the target cumulative descriptor to obtain the descriptor correlation coefficient corresponding to each sampling point. Continuing with the example of neural activity monitoring, the sampling point signal descriptor contains complete information such as the original neural impulse frequency, amplitude, and conduction velocity, and the target instruction descriptor contains all the expected requirements for these features. The computer system merges them to form the control cumulative descriptor, a process similar to the operation of merging component descriptors described above, but here the entire sampling point signal descriptor and the target instruction descriptor are merged. Then, the ratio between the control cumulative descriptor and the target cumulative descriptor is calculated. For example, if the control cumulative descriptor has a value of 80 (assuming a numerical value) through some calculation, and the value of the target cumulative descriptor is 100, then the ratio between them is 0.8, which is the descriptor correlation coefficient corresponding to the sampling point.

[0132] To achieve this step, the computer system can employ an algorithm of data merging and scale calculation. For data merging, the sample point signal descriptor and the target instruction descriptor can be combined together in a similar structural combination manner as before. When calculating the scale, a simple division operation is used to obtain the scale relationship between the two cumulative descriptors, thereby obtaining the descriptor correlation coefficient.

[0133] As an embodiment, step S400, merging the signal descriptor and the instruction descriptor according to the descriptor correlation coefficient to obtain the signal-instruction integrated descriptor, can include:

[0134] Step S410: weighting the sample point signal descriptor according to the descriptor correlation coefficient to obtain the basic signal-instruction integrated descriptor corresponding to each sample point;

[0135] Step S420: merging the basic signal-instruction integrated descriptors of the same feature axis to obtain the signal-instruction integrated descriptor of each feature axis.

[0136] In step S410, the computer system weights the sample point signal descriptor according to the descriptor correlation coefficient, thereby obtaining the basic signal-instruction integrated descriptor corresponding to each sample point. Taking the implantable heart monitoring sensor as an example, it is assumed that the signal descriptor contains information such as heart rate, electrocardiogram waveform features, and the instruction descriptor contains relevant instructions for paying attention to the normal activity range and abnormal state of the heart. At a certain sample point, the descriptor correlation coefficient can reflect the degree of correlation between the signal descriptor and the instruction descriptor. If the descriptor correlation coefficient is high at this sample point, it indicates that the information in the signal descriptor has a strong correlation with the requirements in the instruction descriptor. For example, the heart rate value in the signal descriptor is within the normal heart rate range specified by the instruction descriptor, and the electrocardiogram waveform feature also meets the requirements of the normal heart activity waveform in the instruction descriptor. At this time, the descriptor correlation coefficient can be 0.8.

[0137] The computer system weights the signal descriptor of the sampling point according to the correlation coefficient. For different parts of the signal descriptor, such as the heart rate value part, if the original heart rate value is 70 times / min, the weighted heart rate value obtained by the weighting calculation (assuming the weighting calculation method is the original value multiplied by the correlation coefficient) is 70x0.8=56 times / min (here only to illustrate the weighting operation, the actual meaning may need to be adjusted according to the specific medical meaning and algorithm). Similar weighting operations are also performed on the electrocardiogram waveform feature part, and finally the basic signal instruction integrated descriptor corresponding to the sampling point is obtained. The technical means of such weighting operation can be realized by constructing a weighting function, which takes each element in the original signal descriptor of the sampling point as input, takes the correlation coefficient of the descriptor as a weight factor, and outputs the weighted result, thereby constructing the basic signal instruction integrated descriptor. Subsequently, in step S420, the computer system merges the basic signal instruction integrated descriptors of the same feature axis to obtain the signal instruction integrated descriptor of each feature axis. Continuing to take heart monitoring as an example, assuming that there are multiple sampling points, each of which has obtained its own basic signal instruction integrated descriptor through step S410, and these descriptors are organized according to different feature axes, such as one feature axis corresponding to heart rate related information, another feature axis corresponding to electrocardiogram waveform amplitude related information, etc.

[0138] For the basic signal instruction integrated descriptors of the same feature axis (for example, the feature axis related to heart rate), the computer system merges them. If the heart rate part in the basic signal instruction integrated descriptors of different sampling points is weighted to be 56 times / min (sampling point 1), 60 times / min (sampling point 2), etc. respectively, the computer system will merge these values according to certain merging rules. One possible merging rule is to take the average, so the heart rate value in the signal instruction integrated descriptor on this feature axis may be (56+60+…) / n (n is the number of sampling points). Similar merging operations are also performed on other feature axes (such as electrocardiogram waveform amplitude related feature axes).

[0139] In order to realize such merging operation, the computer system can use data aggregation algorithm. For example, for numerical basic signal instruction integrated descriptors, statistical methods such as summation, averaging, median calculation, etc. can be used for merging. If the basic signal instruction integrated descriptor is in the form of a vector, vector addition followed by normalization can be used to realize the merging, thereby obtaining the signal instruction integrated descriptor of each feature axis.

[0140] As an implementation, step S500, based on the signal instruction integrated descriptor and the signal descriptor, identifying one or more target signal sampling points in the monitoring signal segment, can include:

[0141] Step S510: According to the characteristic axis of the signal descriptor, the signal descriptor is arranged in sequence, and the signal descriptor corresponding to the target characteristic axis is selected from the signal descriptor to obtain the current signal descriptor;

[0142] Step S520: The signal instruction integration descriptor corresponding to the target characteristic axis is selected from the signal instruction integration descriptor to obtain the current signal instruction integration descriptor;

[0143] Step S530: Based on the current signal instruction integration descriptor and the current signal descriptor, the target signal descriptor is extracted from the monitoring signal segment, and one or more target signal sampling points in the monitoring signal segment are determined according to the target signal descriptor.

[0144] In step S510, the computer system arranges the signal descriptor in sequence according to the characteristic axis of the signal descriptor, and selects the signal descriptor corresponding to the target characteristic axis from the signal descriptor according to the arrangement result to obtain the current signal descriptor. Taking an implantable neural activity monitoring sensor as an example, the signal descriptor contains multiple characteristic axes, such as the frequency characteristic axis, amplitude characteristic axis and conduction velocity characteristic axis of neural impulse, etc. The computer system first arranges these characteristic axes in a certain predetermined order, for example, according to the importance of the characteristics or the order of data acquisition. Assuming that according to the importance order, the frequency characteristic axis is arranged first, the amplitude characteristic axis is arranged second, and the conduction velocity characteristic axis is arranged last.

[0145] Then, the computer system selects the signal descriptor corresponding to the target characteristic axis according to the arrangement result. If the frequency characteristic of neural impulse is the focus in the current task, the computer system will select the signal descriptor part corresponding to the frequency characteristic axis from the overall signal descriptor as the current signal descriptor. This operation can be realized by index positioning technology, that is, according to the corresponding relationship between the pre-set characteristic axis and the index, the signal descriptor part corresponding to the target characteristic axis is quickly located.

[0146] Next, in step S520, the computer system selects a signal instruction integration descriptor corresponding to the target feature axis in the signal instruction integration descriptor, to obtain a current signal instruction integration descriptor. Still taking the neural activity monitoring as an example, the signal instruction integration descriptor is a comprehensive descriptor that integrates the signal descriptor and the instruction descriptor information. Having determined that the target feature axis is the frequency feature axis, the computer system searches for the part related to the frequency feature axis in the signal instruction integration descriptor. For example, the signal instruction integration descriptor can contain the instruction requirements for the frequency feature (such as the normal range of frequency, the threshold of abnormal fluctuation, etc.) and the integrated results of the frequency-related signal information. The computer system extracts this part of content as the current signal instruction integration descriptor. This can be achieved through a data matching algorithm, that is, by matching the identification of the feature axis or the semantic information of the related feature, to find the corresponding part of the signal instruction integration descriptor.

[0147] Finally, in step S530, the computer system extracts a target signal descriptor from the monitoring signal segment based on the current signal instruction integration descriptor and the current signal descriptor, and determines one or more target signal sampling points in the monitoring signal segment according to the target signal descriptor. For example, in the neural activity monitoring scenario, the current signal instruction integration descriptor contains the instruction requirements for the frequency feature, such as the normal frequency range of 80-120 times / second, and the current signal descriptor contains the actual frequency values of each sampling point. The computer system compares the actual frequency value of each sampling point with the instruction-required frequency range, extracts the signal descriptor part corresponding to the sampling points falling within or close to this range, as the target signal descriptor.

[0148] For determining the target signal sampling point, if the frequency value of a certain sampling point is 100 times / second, it is within the normal frequency range, and then this sampling point can be determined as the target signal sampling point. To achieve such extraction and determination operation, the computer system can use threshold judgment technology. For numerical features, a reasonable threshold range (such as the frequency range described above) is set to screen out the sampling points meeting the threshold requirements. Meanwhile, pattern matching technology can also be used. For some complex signal patterns (such as a specific neural activity pattern composed of multiple features), a pre-defined pattern template is matched with the actual signal to find the matching successful sampling points as the target signal sampling points.

[0149] Through steps S510-S530, the computer system can gradually locate and identify the target signal sampling points in the monitoring signal segment based on the signal instruction integration descriptor and the signal descriptor, which provides key basic data for the subsequent monitoring signal visualization operation based on these sampling points, helps to improve the accuracy and effectiveness of visualization, and thus better meets the monitoring needs of the implantable medical monitoring sensor.

[0150] As one implementation, step S530, extracting the target signal descriptor from the monitored signal segment based on the current signal instruction integration descriptor and the current signal descriptor, may include:

[0151] Step S531: Based on the current signal instruction integration descriptor and the current signal descriptor, optimize the preset target signal descriptor, and determine the optimized target signal descriptor as the preset target signal descriptor;

[0152] Step S532: Jump to the step of selecting the signal descriptor of the target feature axis from the signal descriptors according to the sequential arrangement result and execute it in a loop until each signal descriptor is the current signal descriptor, and obtain the target signal descriptor.

[0153] In step S531, the computer system optimizes the preset target signal descriptor based on the current signal instruction integration descriptor and the current signal descriptor, and determines the optimized target signal descriptor as the preset target signal descriptor. Taking an implantable cardiac monitoring sensor as an example, the preset target signal descriptor may be a signal feature template initially set to identify target signal sampling points, which includes expected features such as heart rate and ECG waveform. The current signal instruction integration descriptor contains comprehensive information that integrates instruction requirements (such as requirements for normal heart rate range and normal ECG waveform morphology) and some signal information. The current signal descriptor is the actual signal descriptor corresponding to the currently interested feature axis (e.g., the heart rate values ​​and related features of each sampling point corresponding to the heart rate feature axis). Assuming that the preset target signal descriptor specifies a normal heart rate range of 60-100 beats / minute, the QRS complex amplitude of the ECG waveform should be within a certain range. The computer system optimizes the preset target signal descriptor based on more precise instructions in the current signal instruction integration descriptor (such as adjusting the normal heart rate range to 70-90 beats / minute due to individual patient differences or special monitoring needs) and the actual signal situation in the current signal descriptor (such as the distribution of the actual sampling point heart rate value, the actual fluctuation of the QRS complex amplitude, etc.).

[0154] To achieve this optimization, the computer system can employ techniques based on statistical data analysis. For example, it can analyze a large amount of historical monitoring data to identify a more reasonable heart rate range and ECG waveform characteristic range under current conditions (such as a specific patient group or a specific monitoring period). It can also use adaptive algorithms from machine learning to automatically adjust the parameters in the preset target signal descriptor by integrating the descriptor with the data features in the current signal descriptor based on the current signal instruction. For example, using an adaptive filtering algorithm, the judgment criteria for ECG waveform characteristics can be adjusted based on the noise level and signal characteristic changes in the actual signal, thereby optimizing the preset target signal descriptor. Next, in step S532, the computer system jumps to the step of selecting the signal descriptor with the target feature axis from the signal descriptors according to the sequential arrangement result (i.e., step S510) and executes iteratively until each signal descriptor is the current signal descriptor, thus obtaining the target signal descriptor. Taking cardiac monitoring as an example, initially, the system may focus on the signal descriptor corresponding to the heart rate characteristic axis according to the importance of the characteristic axis (step S510). After optimizing the preset target signal descriptor in step S531, the computer system will return to step S510 and start focusing on the next characteristic axis, such as the signal descriptor corresponding to the electrocardiogram waveform characteristic axis.

[0155] For each characteristic axis corresponding to a signal descriptor, the computer system repeats the optimization operation in step S531, continuously adjusting the target signal descriptor. For example, when focusing on the ECG waveform characteristic axis, the system optimizes the target signal descriptor portion related to the ECG waveform based on the current signal instruction integration descriptor's requirements regarding the ECG waveform (such as the normal morphology of the ST segment, the amplitude requirements of the T wave, etc.) and the actual ECG waveform data in the current signal descriptor. This iterative process continues until all signal descriptors corresponding to all characteristic axes have undergone such optimization operations, at which point the final target signal descriptor is obtained.

[0156] In this cyclical process, the computer system needs to record the optimized target signal descriptor for each feature axis and integrate them effectively. This can be achieved by establishing data structures to store and manage the information for each feature axis. For example, matrices or nested data structures can be used to store information such as the name of each feature axis and the optimized target signal descriptor, so that a complete target signal descriptor can be obtained through final integration.

[0157] As one implementation, step S531, optimizing the preset target signal descriptor based on the current signal instruction integration descriptor and the current signal descriptor, may include:

[0158] Step S5311: Based on the current signal instruction integration descriptor and the current signal descriptor, optimize the preset target signal descriptor to obtain the basic target signal descriptor, and determine the basic target signal descriptor as the preset target signal descriptor;

[0159] Step S5312: Jump to the step of integrating the current signal instruction and the current signal descriptor to optimize the preset target signal descriptor, and execute the step repeatedly until the preset number of optimizations is met, and obtain the optimized target signal descriptor.

[0160] In step S5311, the computer system optimizes the preset target signal descriptor based on the current signal instruction integration descriptor and the current signal descriptor to obtain a basic target signal descriptor, which is then determined as the preset target signal descriptor. Taking an implantable blood glucose monitoring sensor as an example, the preset target signal descriptor may include a preliminary setting of the blood glucose concentration range, such as a normal blood glucose concentration range preset to 70-130 mg / dL, as well as some basic requirements for the rate of change of blood glucose concentration. The current signal instruction integration descriptor includes comprehensive information that integrates detection instructions (such as more precise blood glucose monitoring requirements for specific patients, possibly adjusted for factors such as patient age and health status) and some signal information (such as some statistical characteristics of recent blood glucose monitoring data). The current signal descriptor is the actual signal descriptor corresponding to the blood glucose-related feature axis of current interest, such as information including the actual blood glucose concentration values ​​at each sampling point and their changing trends. The computer system performs adaptive weight allocation on the sub-target signal descriptors in the preset target signal descriptor to obtain candidate target signal descriptors. Assume that the sub-target signal descriptors in the preset target signal descriptor are the blood glucose concentration range and the rate of change of blood glucose concentration. Regarding the blood glucose concentration range, if the current signal command descriptor indicates that the patient's normal blood glucose concentration range may be closer to 80-120 mg / dL due to recent dietary adjustments, and the actual blood glucose concentration in the current signal descriptor mostly falls within this range, the computer system will assign a higher weight to this sub-target signal descriptor. Regarding the blood glucose concentration change rate, if the actual monitored blood glucose concentration change is relatively stable and conforms to the preset change rate requirement, it will also be assigned a corresponding weight based on its degree of conformity. This adaptive weight allocation yields candidate target signal descriptors.

[0161] Next, the computer system performs cross-modal weighting on the candidate target signal descriptor, the current signal instruction integrated descriptor, and the current signal descriptor to obtain the current target signal descriptor. Since the candidate target signal descriptor is optimized from the original preset target signal descriptor, it differs in information expression modally from the current signal instruction integrated descriptor and the current signal descriptor. For example, the candidate target signal descriptor describes blood glucose characteristics from the perspective of the original settings, the current signal instruction integrated descriptor contains comprehensive information of instructions and some signals, and the current signal descriptor is the actual blood glucose monitoring signal characteristic. The computer system performs cross-modal weighting based on the correlation and importance among the three. For instance, under certain circumstances, the instruction portion of the current signal instruction integrated descriptor may have a stronger guiding influence on the overall optimization and may be assigned a higher weight; while the actual blood glucose concentration value in the current signal descriptor, because it directly reflects the monitoring situation, will also be assigned a correspondingly important weight. Then, the three are weighted and combined to obtain the current target signal descriptor.

[0162] Finally, the computer system projects the current target signal descriptor onto a preset target representation domain to obtain the basic target signal descriptor. This preset target representation domain is a specific space used to uniformly describe and compare the characteristics of blood glucose monitoring signals. For example, it may be a multi-dimensional space that comprehensively considers factors such as blood glucose concentration, rate of change, and time. Projecting the current target signal descriptor onto this space means re-representing it within this specific spatial framework, thereby obtaining the basic target signal descriptor. This projection operation can be achieved through techniques such as coordinate transformation, for example, converting each element in the current target signal descriptor into coordinate values ​​in the preset target representation domain according to a preset mapping relationship.

[0163] In step S5312, the computer system jumps to the step of integrating the descriptor based on the current signal instruction and optimizing the preset target signal descriptor based on the current signal descriptor (i.e., step S5311) and executes it cyclically until the preset number of optimizations is met, obtaining the optimized target signal descriptor. Continuing with the blood glucose monitoring example, assume the preset number of optimizations is 3. After the first execution of step S5311, a basic target signal descriptor is obtained, but this result may not be ideal. The computer system executes step S5311 again, re-integrating the descriptor based on the new current signal instruction (because as monitoring progresses, this descriptor may contain more new information) and the current signal descriptor to optimize the basic target signal descriptor (which now becomes the new preset target signal descriptor).

[0164] Each iteration adjusts the weighting, cross-modal weighting, and projection operations of the sub-target signal descriptor based on new information. For example, as more blood glucose monitoring data is incorporated into the current signal descriptor, the previously set weights may become inappropriate and require readjustment. After three such iterations, the final target signal descriptor is a fully optimized result that more accurately reflects the blood glucose characteristics required to match the current monitoring situation, laying the foundation for accurate identification of target signal sampling points in the future.

[0165] To achieve these operations, computer systems can employ various techniques. Adaptive weight allocation can utilize self-attention mechanisms or data-driven algorithms, such as determining weights based on the deviation between actual data and preset values, or the distribution of the data. For cross-modal weight allocation, cross-attention mechanisms can be used, or a modal correlation model can be constructed to allocate weights by analyzing the correlation and importance between different modal information. When projecting onto a preset target representation domain, mathematical tools such as linear transformations and nonlinear mappings can be used. These tools can be selected and adjusted according to the specific structure and requirements of the preset target representation domain.

[0166] As one implementation, the preset target signal descriptor includes one or more sub-target signal descriptors. Step S5311, based on the current signal instruction integration descriptor and the current signal descriptor, optimizes the preset target signal descriptor to obtain a basic target signal descriptor, which may include:

[0167] Step S53111: Perform adaptive weight allocation on the sub-target signal descriptors to obtain candidate target signal descriptors;

[0168] Step S53112: Perform cross-modal weight allocation on the candidate target signal descriptor, the current signal command integrated descriptor, and the current signal descriptor to obtain the current target signal descriptor;

[0169] Step S53113: Project the current target signal descriptor onto the preset target representation domain to obtain the basic target signal descriptor.

[0170] In step S53111, the computer system performs adaptive weight allocation on the sub-target signal descriptors to obtain candidate target signal descriptors. Taking an implantable ECG monitoring sensor as an example, the preset target signal descriptor may include multiple sub-target signal descriptors, such as expected descriptions of the QRS complex amplitude, ST segment offset, and heart rate of the ECG signal. Assume the normal range for QRS complex amplitude is preset to 0.5~2.0mV, the normal range for ST segment offset is preset to -0.05~0.05mV, and the normal range for heart rate is preset to 60~100 beats / minute.

[0171] The current signal command integrated descriptor contains comprehensive information that integrates detection commands (such as more precise requirements for specific patients or special monitoring scenarios, perhaps because the patient is taking medications that affect cardiac function, and the focus on QRS amplitude is more stringent, with the normal range adjusted to 0.8~1.8mV) and some signal information (such as recent ECG monitoring data showing that QRS amplitude is mostly between 0.9~1.7mV, ST segment offset is relatively stable between -0.03~0.03mV, and heart rate is mainly concentrated between 70~90 beats / minute). The current signal descriptor is the actual signal descriptor corresponding to the ECG-related characteristic axis of current interest, such as information including the actual QRS amplitude value, ST segment offset value, and heart rate value at each sampling point.

[0172] The computer system performs adaptive weight allocation based on this information. For the QRS group amplitude sub-target signal descriptor, since its actual value is close to the adjusted normal range and falls within the scope of interest, and its distribution in the current signal descriptor indicates its stability, it may be assigned a higher weight, denoted as [missing information]. For the ST segment offset quantum target signal descriptor, although it is relatively stable, its importance is slightly lower than that of the QRS group amplitude, and weighting is assigned accordingly. For the heart rate sub-target signal descriptor, weights are assigned. This weighting can be based on the standard deviation formula of the data distribution. To determine, among which For the values ​​of each sampling point, Let n be the average value and n be the number of sampling points. The standard deviation of the data corresponding to each sub-target signal descriptor is calculated, and the weight is determined based on the relationship between the standard deviation and the overall data; the smaller the standard deviation, the higher the weight. Then, each sub-target signal descriptor is multiplied by its corresponding weight to obtain the candidate target signal descriptors. For example, for the QRS group amplitude part, if the original description is... The weighted description is as follows: This process is repeated for other sub-target signal descriptors, and finally, the candidate target signal descriptors are obtained by combining them.

[0173] Next, in step S53112, the computer system performs cross-modal weight allocation on the candidate target signal descriptor, the current signal instruction integrated descriptor, and the current signal descriptor to obtain the current target signal descriptor. Since the candidate target signal descriptor is optimized from the original preset target signal descriptor, it differs from the current signal instruction integrated descriptor and the current signal descriptor in its information representation. For example, the candidate target signal descriptor describes the ECG characteristics from the perspective of the original settings, the current signal instruction integrated descriptor contains comprehensive information of the instruction and some signals, and the current signal descriptor represents the actual ECG monitoring signal characteristics.

[0174] The result of the target signal descriptor selected by the dummy device after step S53111 is B_1 (corresponding to the QRS group amplitude weighting). (After weighting the corresponding ST segment offset) (After heart rate weighting), the current signal-instruction integrated descriptor is C, and the current signal descriptor is D. The computer system assigns cross-modal weights based on the correlation and importance among the three. If analysis reveals that the instruction portion of the current signal-instruction integrated descriptor C has a stronger guiding influence on overall optimization, it may be assigned a higher weight. The actual ECG data in the current signal descriptor D, because it directly reflects the monitoring situation, will also be assigned significant weight. Then, the weights of the candidate target signal descriptors .

[0175] Then, the current target signal descriptor E is obtained through weighted combination, for example... The weight allocation here can be determined based on the concept of information entropy, the formula for which is: ,in For the event The probability of occurrence. By calculating the information entropy of different modalities, the weights are determined based on the magnitude of the information entropy; the higher the information entropy, the lower the weight.

[0176] Finally, in step S53113, the computer system projects the current target signal descriptor onto a preset target representation domain to obtain a basic target signal descriptor. This preset target representation domain is a specific space used for uniformly describing and comparing the characteristics of ECG monitoring signals. For example, it may be a multi-dimensional space that comprehensively considers factors such as ECG signal amplitude, time, and frequency.

[0177] Projecting the current target signal descriptor E into this space means re-representing it within this specific spatial framework. Assume the preset target representation domain is a three-dimensional space, corresponding to the ECG signal amplitude x, time t, and frequency f, respectively. The computer system, based on a pre-defined mapping relationship F, for example... The process involves converting each element of the current target signal descriptor E into coordinate values ​​(x, y, z) in a predefined target representation domain, thus obtaining the basic target signal descriptor. This mapping relationship can be implemented based on linear transformations or nonlinear mapping functions; for example, matrix multiplication can be used for linear transformations. ,in Let A be the input vector (elements in the current target signal descriptor) and A be the transformation matrix. The output vector is the coordinate value in the predefined target representation domain; for nonlinear mappings, activation functions such as the sigmoid function in neural networks can be used. Methods such as ) are used to construct mapping relationships and convert the current target signal descriptor into the basic target signal descriptor.

[0178] As one implementation, step S200, implicitly representing the monitored signal segment to obtain a signal descriptor with one or more characteristic axes, may include:

[0179] Step S210: Perform multi-level implicit representation on the monitored signal segment to obtain one or more feature-level basic signal descriptors;

[0180] Step S220: Normalize the basic signal descriptor to obtain one or more candidate signal descriptors with characteristic axes;

[0181] Step S230: Based on the preset feature fusion path, perform descriptor fusion on the candidate signal descriptors to obtain signal descriptors with one or more feature axes.

[0182] In step S210, the computer system performs multi-level (i.e., scale-based) implicit representation on the monitored signal segment to obtain one or more basic signal descriptors at the feature level. Taking an implanted electroencephalogram (EEG) monitoring sensor as an example, the monitored signal segment is EEG signal data. EEG signals have complex spatiotemporal characteristics; different frequency components and spatial distributions reflect different neural activity states of the brain. The computer system uses multi-level implicit representation to capture these characteristics.

[0183] For example, on a time scale, EEG signals can be decomposed into different frequency bands, such as... Frequency band (0-4Hz) Frequency band (4-8Hz) Frequency band (8-13Hz) Frequency band (13-30Hz) and Frequency bands (above 30Hz) represent different levels of implicit representation. Spatially, the scalp can be partitioned based on electrode placement, with the characteristics of the EEG signal in each partition at different frequency bands serving as a basic signal descriptor for a feature level. This decomposition can be achieved using techniques such as wavelet transform. The wavelet transform formula is... , where x(t) is the original signal (EEG signal). Here, 'a' is a wavelet function, 'a' is a scaling factor, and 'b' is a translation factor. By choosing different values ​​for 'a' and 'b', signal representations at different scales can be obtained, and these representations constitute the basic signal descriptors for different feature levels.

[0184] Next, in step S220, the computer system normalizes the basic signal descriptors to obtain one or more candidate signal descriptors for characteristic axes. Continuing with the EEG signal as an example, after obtaining the basic signal descriptors at different feature levels, normalization processing is required because signals from different frequency bands and spatial partitions may differ in numerical range, feature importance, etc.

[0185] For example, for a basic signal descriptor at a certain feature level, if it is The original amplitude values ​​of EEG signals in a specific spatial region within a frequency band may have a large and non-uniform range. The computer system selects target basic signal descriptors (such as signals from a specific frequency band and spatial region) from the basic signal descriptors based on feature hierarchy, and adjusts the dimension of the target basic signal descriptor. Principal component analysis (PCA) can be used to adjust the dimension. The goal of PCA is to find the principal components in the data, so that the data retains as much of the original information as possible in the new low-dimensional space. Let the original data matrix X have a covariance matrix of... Eigenvalue decomposition of C ,in V is a diagonal matrix, with diagonal elements being eigenvalues, and V is the eigenvector matrix. The projection matrix P is formed by selecting the eigenvectors corresponding to the k largest eigenvalues. Then, the low-dimensional data Y = XP, where Y is the signal descriptor after adjusting the dimension.

[0186] The adjusted signal descriptor is then combined with the fixed-position code to obtain a combined signal descriptor, which includes one or more sampled-point signal descriptors. The fixed-position code can be encoding information related to the location of the signal sampling points, such as assigning a specific encoding value to each sampling point according to time or spatial order. Assume the adjusted signal descriptor is... (i represents the sampling point index), fixed position encoding is Then the combined signal descriptor .

[0187] Finally, adaptive weight allocation is performed on the sampled point signal descriptors to obtain weighted signal descriptors. The base signal descriptors are then optimized based on these weighted descriptors to obtain candidate signal descriptors with one or more characteristic axes. The adaptive weight allocation can be determined based on certain statistical properties of the signal, such as its energy or variance. Let the sampled point signal descriptors be... Calculate weights based on energy ,in express Energy (which can be obtained through) (Calculated by the sum of squares), where n is the number of sampling points. Weighted signal descriptor Then, by integrating and optimizing all weighted sample point signal descriptors, one or more candidate signal descriptors for characteristic axes are obtained.

[0188] Finally, in step S230, the computer system performs descriptor fusion on the candidate signal descriptors based on a preset feature fusion path to obtain signal descriptors with one or more feature axes. Taking an EEG signal as an example, assume there are multiple candidate signal descriptors, corresponding to different frequency bands and different spatial partitions, after processing in the previous steps. The preset feature fusion path includes a first feature fusion path and a second feature fusion path.

[0189] First, based on the descriptor dimension of the candidate signal descriptors, the candidate signal descriptors are arranged in order to obtain the first dimension ordered arrangement result. For example, the candidate signal descriptors are arranged in ascending order of descriptor dimension. Then, based on the first dimension ordered arrangement result, the candidate signal descriptors are fused according to the first feature fusion path to obtain one or more candidate fused descriptors.

[0190] The first feature fusion path can be a step-by-step merging approach. For example, among the candidate signal descriptors arranged in order, the candidate signal descriptor with the smallest descriptor dimension is selected to obtain the current candidate signal descriptor, and the current candidate signal descriptor is determined as the first candidate fusion descriptor. Let the first candidate signal descriptor be... This is the first candidate fusion descriptor. Based on the ordered arrangement of the first dimension, the next candidate signal descriptor is selected from the candidate signal descriptors, thus obtaining the first pseudo-fusion descriptor of the current candidate signal descriptor. Assume... yes The next alternative signal descriptor is then for The first pseudo-fusion descriptor.

[0191] The current candidate signal descriptor is merged with the first quasi-fusion descriptor to obtain the second candidate fusion descriptor, and the first quasi-fusion descriptor is determined as the current candidate signal descriptor. The merging method can be direct concatenation or weighted summation, etc. For example, if direct concatenation is used, let... The dimension is m. If the dimension is n, then the second alternative fusion descriptor The dimension is m+n. This process is then repeated until all candidate signal descriptors are fused, resulting in one or more candidate fused descriptors.

[0192] Next, based on the descriptor dimension of the candidate fusion descriptors, the candidate fusion descriptors are arranged in order to obtain the second-dimensional ordered arrangement result. Then, based on the second-dimensional ordered arrangement result, the candidate fusion descriptors are fused according to the second feature fusion path to obtain one or more fusion descriptors, and each fusion descriptor is determined as a signal descriptor for a feature axis. The second feature fusion path can be based on the importance of features. For example, among the ordered candidate fusion descriptors, the candidate fusion descriptor with the largest descriptor dimension is selected to obtain the current fusion descriptor, and the features of different descriptor components in the current fusion descriptor are merged. Let the current fusion descriptor be... The features of its different frequency bands, spatial partitions and other components can be merged according to certain rules (such as weighted average) to obtain the first fusion descriptor.

[0193] Based on the ordered arrangement of the second dimension, the next candidate fusion descriptor for the current fusion descriptor is selected from the candidate fusion descriptors, thus obtaining the second pseudo-fusion descriptor for the current fusion descriptor. Assume... yes The next alternative fusion descriptor is then for The second pseudo-fusion descriptor is then used. The current fusion descriptor and the second pseudo-fusion descriptor are merged to obtain the second fusion descriptor, and the second pseudo-fusion descriptor is determined as the current fusion descriptor. For example, a weighted summation method can be used for merging, with weights set as follows: and Then the second fusion descriptor This process is repeated until all candidate fusion descriptors have been fused, resulting in one or more fused descriptors, which are signal descriptors for one or more characteristic axes.

[0194] As one implementation, step S220, normalizing the basic signal descriptor to obtain one or more candidate signal descriptors with characteristic axes, may include:

[0195] Step S221: Based on the feature level, select the target basic signal descriptor from the basic signal descriptors, adjust the dimension of the target basic signal descriptor, and obtain the adjusted signal descriptor;

[0196] Step S222: Combine the adjustment signal descriptor with the fixed position code to obtain a combined signal descriptor, which includes one or more sampling point signal descriptors;

[0197] Step S223: Perform adaptive weight allocation on the sampled point signal descriptors to obtain weighted signal descriptors, and optimize the basic signal descriptors based on the weighted signal descriptors to obtain one or more candidate signal descriptors for characteristic axes.

[0198] In step S221, the computer system selects a target basic signal descriptor from the basic signal descriptors according to the feature hierarchy, and adjusts the dimension of the target basic signal descriptor. Taking an implantable electrocardiogram (ECG) monitoring sensor as an example, the basic signal descriptor may contain information about the ECG signal in different frequency bands, different time intervals, and different leads (which can be regarded as a spatial distribution feature), which constitute different feature levels.

[0199] Suppose we are interested in the basic signal descriptor of ECG signals in a specific frequency band (e.g., alpha band, 8-13 Hz) and a specific lead (e.g., lead V1), the computer system selects this as the target basic signal descriptor. This target basic signal descriptor may have a high dimension, containing much redundant information or noise components. To adjust its dimension, the computer system can employ various techniques, such as principal component analysis (PCA).

[0200] Let X be the matrix composed of the target basic signal descriptors, with a size of m×n (m is the number of samples, and n is the number of features, which is the dimension). First, calculate the covariance matrix. Then perform eigenvalue decomposition on C. ,in It is a diagonal matrix, and the elements on the diagonal are the eigenvalues. V is the eigenvector matrix. The eigenvectors are sorted in descending order of eigenvalues, and the eigenvectors corresponding to the k largest eigenvalues ​​are selected to form the projection matrix P (P is n×k). Then, the signal descriptor Y=XP after adjusting the dimension has a new dimension of k. In this way, redundant information is reduced and the main features are retained.

[0201] Next, in step S222, the computer system combines the modulated signal descriptor with the fixed-position code to obtain a combined signal descriptor, which includes one or more sample point signal descriptors. Continuing with the above example of ECG monitoring, the modulated signal descriptor Y contains characteristic information of the ECG signal in a specific frequency band and lead after dimensionality adjustment. The fixed-position code can be a code related to the position of the ECG signal sampling point in the time series.

[0202] For example, assuming an electrocardiogram (ECG) signal is sampled at fixed time intervals, each sample point can be assigned a sequential number as a fixed-position code. If each row in the modulated signal descriptor Y corresponds to a feature vector of one sample point, let... Let be the i-th row vector in Y (representing the adjusted signal characteristics of the i-th sampling point). Encode the fixed position of the i-th sampling point (e.g.) Then the combined signal descriptor This combines location information with signal feature information to form a combined signal descriptor that contains one or more sampling point signal descriptors.

[0203] Finally, in step S223, the computer system adaptively assigns weights to the sampled point signal descriptors to obtain weighted signal descriptors. Based on these weighted signal descriptors, the basic signal descriptors are optimized to obtain candidate signal descriptors for one or more characteristic axes. Taking ECG monitoring as an example again, for combined signal descriptors… The computer system performs adaptive weight allocation according to certain rules.

[0204] One possible approach is to assign weights based on the energy of the signal. Its energy can be defined as (in yes The j-th element, n is (Dimension of). Let the total energy be... (where m is the number of sampling points), then the weight of sampling point i Weighted signal descriptor .

[0205] The base signal descriptors are optimized based on these weighted signal descriptors. Information from the weighted signal descriptors can be reintegrated into the structure of the base signal descriptors in a specific way. For example, if the base signal descriptors were initially organized according to different frequency bands and leads, then based on the weights of different sampling points in the weighted signal descriptors, weighted summation or other forms of integration operations can be performed on the base signal descriptors for each frequency band and lead to obtain one or more candidate signal descriptors for each characteristic axis. These candidate signal descriptors are more standardized in structure and content than the original base signal descriptors, and can be better used for subsequent signal descriptor fusion operations, facilitating the visualization of monitoring signals.

[0206] As one implementation, the preset feature fusion path includes a first feature fusion path and a second feature fusion path. Step S230, based on the preset feature fusion path, performs descriptor fusion on the candidate signal descriptors to obtain signal descriptors with one or more feature axes, which may include:

[0207] Step S231: Arrange the candidate signal descriptors in order according to their descriptor dimensions to obtain the first dimension ordered arrangement result;

[0208] Step S232: Based on the first dimension ordered arrangement result, perform descriptor fusion on the candidate signal descriptors according to the first feature fusion path to obtain one or more candidate fused descriptors;

[0209] Step S233: Arrange the candidate fusion descriptors in order according to their descriptor dimensions to obtain the second dimension ordered arrangement result;

[0210] Step S234: Based on the ordered arrangement result of the second dimension, perform descriptor fusion on the candidate fusion descriptors according to the second feature fusion path to obtain one or more fusion descriptors, and determine each fusion descriptor as a signal descriptor with a feature axis.

[0211] In step S231, the computer system arranges the candidate signal descriptors in order according to their descriptor dimensions to obtain the first-dimensional ordered arrangement result. Taking an implantable electroencephalogram (EEG) monitoring sensor as an example, the candidate signal descriptors may be obtained from EEG signals from different brain regions or different frequency bands after processing in the previous steps. These candidate signal descriptors have different dimensions, reflecting the complexity of the information they contain or the number of features.

[0212] For example, suppose the dimension of candidate signal descriptor A obtained from the frontal lobe is 3, the dimension of candidate signal descriptor B obtained from the occipital lobe is 5, and the dimension of candidate signal descriptor C obtained from the alpha band is 4. The computer system arranges them in ascending order of dimension, resulting in the first dimension ordered arrangement, i.e., A, C, B. This operation can be achieved by comparing the dimensions of each candidate signal descriptor and then using a sorting algorithm (such as bubble sort). The basic idea of ​​the bubble sort algorithm is to compare adjacent elements sequentially, swapping them if they are in the wrong order. After multiple rounds of comparisons and swaps, the entire sequence is eventually arranged in an ordered manner.

[0213] Next, in step S232, based on the ordered arrangement result of the first dimension, the computer system performs descriptor fusion on the candidate signal descriptors according to the first feature fusion path to obtain one or more candidate fused descriptors. Continuing with the EEG signal as an example, the first feature fusion path can adopt a step-by-step merging strategy.

[0214] First, among the candidate signal descriptors arranged in order, the computer system selects the candidate signal descriptor with the smallest descriptor dimension to obtain the current candidate signal descriptor, and determines it as the first candidate fusion descriptor. Following the previous example, A has the smallest dimension, so A is determined as the first candidate fusion descriptor.

[0215] Then, based on the ordered arrangement of the first dimension, the computer system selects the next candidate signal descriptor from the candidate signal descriptors, thus obtaining the first pseudo-fusion descriptor of the current candidate signal descriptor. For A, its next candidate signal descriptor is C, so C is the first pseudo-fusion descriptor of A.

[0216] The current candidate signal descriptor is merged with the first quasi-fusion descriptor to obtain the second candidate fusion descriptor, and the first quasi-fusion descriptor is determined as the current candidate signal descriptor. Assuming a direct splicing method is used for merging, if... , Then the second alternative fusion descriptor Meanwhile, C is selected as the current candidate signal descriptor.

[0217] Next, the computer system jumps to the step of selecting the next candidate signal descriptor from the candidate signal descriptors based on the first dimension's ordered arrangement results, and repeats this process until all candidate signal descriptors are fused, resulting in one or more candidate fused descriptors. For example, in the next iteration, if the current candidate signal descriptor is C (the first proposed fused descriptor mentioned earlier), and its next candidate signal descriptor is B, then C and B are merged according to the previous merging method to obtain a new candidate fused descriptor. Through this iterative operation, one or more candidate fused descriptors are ultimately obtained. In step S233, the computer system arranges the candidate fused descriptors in order based on their descriptor dimensions to obtain the second dimension's ordered arrangement results. For example, after step S232, two candidate fused descriptors D and E are obtained. Assuming D has a dimension of 7 and E has a dimension of 9, they are arranged in ascending order of dimension, resulting in the second dimension's ordered arrangement results as D and E. This operation can also be implemented using a sorting algorithm. Finally, in step S234, based on the ordered arrangement of the second dimension, the computer system performs descriptor fusion on the candidate fusion descriptors according to the second feature fusion path to obtain one or more fusion descriptors, and determines each fusion descriptor as a signal descriptor for a feature axis. The second feature fusion path can be based on the importance of the features.

[0218] Taking EEG signals as an example, among the candidate fusion descriptors arranged in order, the computer system selects the candidate fusion descriptor with the largest descriptor dimension to obtain the current fusion descriptor. Then, it merges the features of different descriptor components in the current fusion descriptor to obtain the first fusion descriptor. Assume that E has the largest dimension and is the current fusion descriptor. If E contains feature components from different brain regions and frequency bands, such as... ,in It may be related to the frontal lobe region. Associated with the occipital lobe brain region, and Frequency band correlation. The computer system assigns pre-defined importance weights to each brain region and frequency band (e.g., a weight of 0.3 for the frontal lobe and 0.4 for the occipital lobe). The frequency band weight is 0.3. These components are weighted and merged to obtain the first fused descriptor. Then, based on the ordered arrangement of the second dimension, the computer system selects the next candidate fused descriptor from the candidate fused descriptors to obtain the second pseudo-fused descriptor of the current fused descriptor. For E, its next candidate fused descriptor is D, so D is the second pseudo-fused descriptor of E. The current fused descriptor and the second pseudo-fused descriptor are merged to obtain the second fused descriptor, and the second pseudo-fused descriptor is determined as the current fused descriptor. Assuming that a weighted summation method is used for merging, let the weight of E be... The weight of D is If E is obtained after the previous processing D is obtained after the previous processing. Then the second fusion descriptor Meanwhile, D is determined as the current fusion descriptor.

[0219] The computer system continuously jumps to the step of selecting the next candidate fusion descriptor from the candidate fusion descriptors based on the ordered results of the second dimension, and repeats this process until all candidate fusion descriptors have been fused, resulting in one or more fusion descriptors. These fusion descriptors are signal descriptors with one or more characteristic axes. They integrate the feature information of different candidate signal descriptors and are optimized and integrated according to a specific fusion path, providing a more effective data representation for subsequent monitoring signal analysis and visualization.

[0220] As one implementation method, step S232, based on the ordered arrangement result of the first dimension, involves fusing the candidate signal descriptors according to the first feature fusion path to obtain one or more candidate fused descriptors, which may include:

[0221] Step S2321: Select the candidate signal descriptor with the smallest descriptor dimension from the candidate signal descriptors, obtain the current candidate signal descriptor, and determine the current candidate signal descriptor as the first candidate fusion descriptor;

[0222] Step S2322: Based on the ordered arrangement of the first dimension, select the next candidate signal descriptor from the candidate signal descriptors to obtain the first pseudo-fusion descriptor of the current candidate signal descriptor;

[0223] Step S2323: Merge the current candidate signal descriptor with the first quasi-fusion descriptor to obtain the second candidate fusion descriptor, and determine the first quasi-fusion descriptor as the current candidate signal descriptor;

[0224] Step S2324: Jump to the step of selecting the next candidate signal descriptor from the candidate signal descriptors based on the results arranged in order according to the first dimension, and repeat the process until all candidate signal descriptors are fused to obtain one or more candidate fused descriptors, including the first candidate fused descriptor and the second candidate fused descriptor.

[0225] In step S2321, the computer system selects the candidate signal descriptor with the smallest descriptor dimension from the candidate signal descriptors, obtains the current candidate signal descriptor, and determines the current candidate signal descriptor as the first candidate fusion descriptor. Taking an implantable cardiac monitoring sensor as an example, assume there are three candidate signal descriptors, A, B, and C, where A has a dimension of 3, B has a dimension of 5, and C has a dimension of 4. By comparing the dimensions of these three candidate signal descriptors, the computer system determines that A is the candidate signal descriptor with the smallest descriptor dimension, and thus selects A as the current candidate signal descriptor and determines it as the first candidate fusion descriptor. This operation of comparing the dimensions can be implemented using a simple numerical comparison algorithm, such as setting up a loop in the program to traverse the dimension of each candidate signal descriptor and record the minimum value and its corresponding candidate signal descriptor. Next, in step S2322, based on the ordered arrangement of the first dimension, the computer system selects the next candidate signal descriptor from the candidate signal descriptors, thus obtaining the first pseudo-fusion descriptor of the current candidate signal descriptor. Following the previous example, since A is the current candidate signal descriptor, in the order arranged from smallest to largest dimension, the next candidate signal descriptor after A is C. Therefore, the computer system selects C as the first pseudo-fusion descriptor of A. This operation is a sequential search based on the previously completed ordered arrangement of the first dimension; in program implementation, the next candidate signal descriptor can be located using an index or pointer. Then, in step S2323, the computer system merges the current candidate signal descriptor with the first pseudo-fusion descriptor to obtain the second candidate fusion descriptor. Assume... , One approach to merging is to perform descriptor component transformation (channel transformation). Let's assume this channel transformation expands the dimension to the same preset number of components. Assuming the preset number of components is 5, for A, the descriptor component transformation is performed by converting it to a value using a mapping function f (e.g., a linear interpolation function, if it's necessary to maintain certain numerical relationships during dimension expansion). ,here and It is based on and The new value obtained through the mapping function f makes the number of components of A' become 5.

[0226] Next, the descriptor dimension of A' is upsampled. Upsampling can be performed using an interpolation algorithm, such as linear interpolation. Assuming a simple linear interpolation algorithm is used, if we want to upsample the dimension of A' from 3 (the original effective dimension) to the same dimension 4 as C (assuming the dimension of C is the target dimension), for each element in A'... new value It can be done through formula (When i=1,2,3). We calculate and obtain the upsampled signal descriptor A'', whose descriptor dimension is the same as C. Simultaneously, we perform a descriptor component transformation on C to obtain the first pseudo-fused descriptor after transformation with a preset number of components. Assuming we perform the same component transformation of C using the same mapping function f (this is just for illustration; in practice, different component transformation methods may be used depending on the specific situation), we obtain... .

[0227] Finally, the upsampled signal descriptor A'' is combined with the transformed first quasi-fused descriptor C' to obtain the second candidate fused descriptor. For example, the combination method can be a simple vector concatenation to obtain the second candidate fused descriptor. .

[0228] In step S2324, the computer system jumps to the step of selecting the next candidate signal descriptor from the candidate signal descriptors based on the ordered results of the first dimension (i.e., step S2322) and executes this step repeatedly until all candidate signal descriptors have completed descriptor fusion, resulting in one or more candidate fused descriptors. Continuing the previous example, the current candidate signal descriptor becomes C (because C became the current candidate signal descriptor in the previous operation), and its next candidate signal descriptor is B. The computer system merges C and B according to the operations of steps S2322-S2323 to obtain a new candidate fused descriptor. This process is repeated until all candidate signal descriptors have participated in the fusion operation, ultimately obtaining one or more candidate fused descriptors. These candidate fused descriptors contain the feature information of the different candidate signal descriptors and are integrated according to specific fusion rules, providing a basis for subsequent further fusion operations (such as the operation in step S234).

[0229] As one implementation, step S2323, merging the current candidate signal descriptor with the first proposed fusion descriptor to obtain the second candidate fusion descriptor, may include:

[0230] Step S23231: Perform descriptor component transformation on the current candidate signal descriptor to obtain a transformed signal descriptor with a preset number of components;

[0231] Step S23232: Upsample the descriptor dimension of the transformed signal descriptor to obtain an upsampled signal descriptor. The descriptor dimension of the upsampled signal descriptor is the same as that of the first pseudo-fused descriptor.

[0232] Step S23233: Perform descriptor component transformation on the first pseudo-fusion descriptor to obtain the transformed first pseudo-fusion descriptor with a preset number of components;

[0233] Step S23234: Combine the upsampled signal descriptor with the transformed first quasi-fusion descriptor to obtain the second alternative fusion descriptor.

[0234] In step S23231, the computer system performs descriptor component transformation (i.e., channel conversion) on the current candidate signal descriptor to obtain a transformed signal descriptor with a preset number of components. Taking an implanted electroencephalogram (EEG) monitoring sensor as an example, assuming that the current candidate signal descriptor A represents the EEG signal characteristics from a specific brain region, it may have n components (e.g., The components here can be understood as characteristic elements describing EEG signals from different perspectives, such as energy values ​​of different frequency bands.

[0235] Given a preset number of components m (m>n), the computer system can employ linear interpolation techniques to perform descriptor component transformation. Assuming a simple linear interpolation method is used, for each component... The computer system calculates the new component values ​​according to the following formula. :

[0236] ;

[0237] (when hour), In this way, A is transformed into a transformed signal descriptor with m components. .

[0238] Next, in step S23232, the computer system upsamples the descriptor dimension of the transformed signal descriptor to obtain an upsampled signal descriptor, and the descriptor dimension of the upsampled signal descriptor is the same as the descriptor dimension of the first pseudo-fused descriptor. Assume that the dimension of the transformed signal descriptor A' obtained after step S23231 is p, and the dimension of the first pseudo-fused descriptor B is q(p). <q)。

[0239] To perform upsampling, computer systems can employ various interpolation algorithms; here, we take bilinear interpolation as an example. Suppose A' is a two-dimensional signal descriptor (e.g., representing EEG signal features in a time-frequency plane). For each element in A'... (where i represents the row index and j represents the column index), the bilinear interpolation formula is:

[0240] ;

[0241] in , , and It is the integer index in A' that is closest to i and j. By calculating each element in A' using this bilinear interpolation algorithm, we obtain the upsampled signal descriptor A'', whose dimension becomes q, which is the same as the dimension of the first pseudo-fusion descriptor B.

[0242] Then, in step S23233, the computer system performs descriptor component transformation on the first pseudo-fusion descriptor to obtain a transformed first pseudo-fusion descriptor with a preset number of components. Assume the first pseudo-fusion descriptor... Similarly, the linear interpolation method in step S23231 is used (this is only to maintain consistency of operation; in fact, other suitable methods can be used depending on the specific situation), and the transformed first pseudo-fusion descriptor is calculated according to a formula similar to that in step S23231. .

[0243] Finally, in step S23234, the computer system combines the upsampled signal descriptor with the transformed first pseudo-fusion descriptor to obtain a second alternative fusion descriptor. For example, if the upsampled signal descriptor... The first pseudo-fusion descriptor after transformation Computer systems can use vector concatenation to combine data.

[0244] Assuming the second alternative fusion descriptor after combination is C, then This completes the merging process from the current candidate signal descriptor and the first proposed fusion descriptor to the second candidate fusion descriptor, laying the foundation for subsequent iterative operations (such as continuing fusion with other candidate signal descriptors in step S2324). This merging operation gradually integrates different signal descriptors through operations such as component transformation, dimensionality upsampling, and combination, which helps to better integrate various feature information in monitoring signal visualization methods.

[0245] As one implementation method, step S234, based on the ordered arrangement result of the second dimension, performs descriptor fusion on the candidate fusion descriptors according to the second feature fusion path to obtain one or more fusion descriptors, which may include:

[0246] Step S2341: Select the candidate fusion descriptor with the largest descriptor dimension from the candidate fusion descriptors to obtain the current fusion descriptor, and merge the features of different descriptor components in the current fusion descriptor to obtain the first fusion descriptor;

[0247] Step S2342: Based on the ordered arrangement of the second dimension, select the next candidate fusion descriptor of the current fusion descriptor from the candidate fusion descriptors, and obtain the second pseudo-fusion descriptor of the current fusion descriptor;

[0248] Step S2343: Merge the current fusion descriptor with the second proposed fusion descriptor to obtain the second fusion descriptor, and determine the second proposed fusion descriptor as the current fusion descriptor;

[0249] Step S2344: Jump to the step of selecting the next candidate fusion descriptor from the candidate fusion descriptors based on the results arranged in order according to the second dimension, and repeat the process until all candidate fusion descriptors have completed descriptor fusion, and one or more fusion descriptors are obtained. The fusion descriptors include the first fusion descriptor and the second fusion descriptor.

[0250] In step S2341, the computer system selects the candidate fusion descriptor with the largest descriptor dimension from the candidate fusion descriptors, obtains the current fusion descriptor, and merges the features of different descriptor components in the current fusion descriptor to obtain the first fusion descriptor. Taking an implantable cardiac monitoring sensor as an example, assume there are three candidate fusion descriptors A, B, and C, with dimensions of 5, 7, and 9, respectively. The computer system determines C as the candidate fusion descriptor with the largest descriptor dimension by comparing their dimensions and selects it as the current fusion descriptor. Assume C contains different descriptor components, such as heart rate feature components. ECG waveform characteristic components To merge these different descriptor component features, computer systems can employ weighted summation techniques. For example, weights can be assigned based on the predetermined importance of each component; if the heart rate feature component... weight ECG waveform characteristic components weight Then the merged eigenvalues The first fused descriptor is obtained by weighted summation of all different descriptor components according to their respective weights.

[0251] Next, in step S2342, based on the ordered arrangement of the second dimension, the computer system selects the next candidate fusion descriptor from the candidate fusion descriptors to obtain the second pseudo-fusion descriptor of the current fusion descriptor. Following the previous example, since C is the current fusion descriptor, in the dimensional arrangement, the candidate fusion descriptor after C is B, so the computer system selects B as the second pseudo-fusion descriptor of C. This operation is based on the previously completed ordered arrangement of the second dimension and a sequential search. Then, in step S2343, the computer system merges the current fusion descriptor with the second pseudo-fusion descriptor to obtain the second fusion descriptor, and determines the second pseudo-fusion descriptor as the current fusion descriptor. Assume the first fusion descriptor (obtained by merging C) is... The second pseudo-fusion descriptor The computer system can merge the components directly to obtain the second fusion descriptor. Meanwhile, B is identified as the current fusion descriptor.

[0252] In step S2344, the computer system jumps to the step of selecting the next candidate fusion descriptor from the candidate fusion descriptors based on the ordered results of the second dimension (i.e., step S2342) and repeats this process until all candidate fusion descriptors have completed fusion, resulting in one or more fusion descriptors. Continuing the previous example, if the current fusion descriptor becomes B and its next candidate fusion descriptor is A, the computer system merges B and A according to steps S2342-S2343 to obtain a new fusion descriptor. This process is repeated until all candidate fusion descriptors have participated in the fusion operation, ultimately resulting in one or more fusion descriptors. These fusion descriptors integrate the feature information of each candidate fusion descriptor and are optimized and integrated according to a specific fusion path. Each fusion descriptor can be identified as a signal descriptor for a feature axis, providing a more effective data representation for subsequent monitoring signal analysis and visualization. The method for determining weights can be based on various factors. For example, statistical analysis can be performed based on a large amount of historical monitoring data; if a certain feature component has a high accuracy rate in judging cardiac health status, it is assigned a higher weight. For direct splicing operations, this is a simple and effective combination method that can combine different descriptors sequentially, facilitating subsequent processing of the overall fused descriptor. This step-by-step fusion approach helps to fully integrate signal feature information from different sources or at different levels, improving the accuracy and effectiveness of monitoring signal visualization.

[0253] This invention provides a computer system, such as... Figure 2As shown, the computer system 100 includes a processor 101 and a memory 103. The processor 101 and the memory 103 are connected, for example, via a bus 102. Optionally, the computer system 100 may also include a transceiver 104. It should be noted that in practical applications, the transceiver 104 is not limited to one type, and the structure of this computer system 100 does not constitute a limitation on the embodiments of the present invention.

Claims

1. A method for visualizing monitoring signals applied to implantable medical monitoring sensors, characterized in that, include: Acquire a target monitoring signal data stream and a detection guidance instruction corresponding to the target monitoring signal data stream, wherein the target monitoring signal data stream includes one or more monitoring signal segments; The monitoring signal segment is implicitly represented to obtain a signal descriptor with one or more characteristic axes, and the detection guidance command is implicitly represented to obtain a command descriptor; Obtain the descriptor correlation coefficient between the signal descriptor and the instruction descriptor at each sampling point, wherein the descriptor correlation coefficient characterizes the degree of correlation between the signal descriptor and the instruction descriptor at the same sampling point; Based on the descriptor correlation coefficient, the signal descriptor and the instruction descriptor are merged to obtain a signal-instruction integrated descriptor; Based on the signal instruction integration descriptor and the signal descriptor, one or more target signal sampling points are identified in the monitored signal segment; Based on one or more target signal sampling points identified, the target monitoring signal data stream is displayed using preset visualization elements to obtain the display monitoring signal.

2. The monitoring signal visualization method for implantable medical monitoring sensors as described in claim 1, characterized in that, The step of obtaining the descriptor correlation coefficient between the signal descriptor and the instruction descriptor at each sampling point includes: Project the signal descriptor and the instruction descriptor onto the same representation domain to obtain the target signal descriptor and the target instruction descriptor corresponding to each feature axis; Extract the sampling point signal descriptor corresponding to each sampling point from the target signal descriptor, and obtain the descriptor correlation coefficient between the sampling point signal descriptor and the target instruction descriptor.

3. The monitoring signal visualization method applied to implantable medical monitoring sensors as described in claim 2, characterized in that, The instruction descriptor includes one or more instruction unit descriptors. Projecting the signal descriptor and the instruction descriptor onto the same representation domain to obtain a target signal descriptor for each feature axis and a target instruction descriptor corresponding to the target signal descriptor includes: The number of descriptor components is determined in the signal descriptor, and the number of signal components is obtained; the number of descriptor components is determined in the instruction descriptor, and the number of instruction components is obtained. Based on the number of signal components and the number of instruction components, the number of target components in the representation domain corresponding to each feature axis is determined, and the instruction unit descriptors are merged to obtain merged instruction descriptors; The number of components of the signal descriptor and the number of components of the merged instruction descriptor are transformed into the target number of components, respectively, to obtain the target signal descriptor for each feature axis and the target instruction descriptor corresponding to the target signal descriptor; The step of obtaining the descriptor correlation coefficient between the sampled point signal descriptor and the target instruction descriptor includes: Extract the descriptor of each descriptor component from the signal descriptor at the sampling point to obtain the signal component descriptor; The descriptor under the descriptor component corresponding to the signal component descriptor is selected from the target instruction descriptor to obtain the instruction component descriptor; Based on the signal component descriptor and the instruction component descriptor, the descriptor correlation coefficient between the sampling point signal descriptor and the target instruction descriptor is determined.

4. The monitoring signal visualization method for implantable medical monitoring sensors as described in claim 3, characterized in that, Determining the descriptor correlation coefficient between the sampled point signal descriptor and the target command descriptor based on the signal component descriptor and the command component descriptor includes: The signal component descriptor and the instruction component descriptor are merged to obtain the merged component descriptor corresponding to each descriptor component; The merged component descriptors of each sampling point are fused to obtain the target cumulative descriptor corresponding to each sampling point; The sampling point signal descriptor is merged with the target instruction descriptor to obtain a control cumulative descriptor, and the ratio between the control cumulative descriptor and the target cumulative descriptor is obtained to obtain the descriptor correlation coefficient corresponding to each sampling point. The step of merging the signal descriptor and the instruction descriptor based on the descriptor correlation coefficient to obtain a signal-instruction integrated descriptor includes: Based on the descriptor correlation coefficient, the sample point signal descriptors are weighted to obtain the basic signal instruction integrated descriptor corresponding to each sample point; The basic signal instruction integrated descriptors of the same characteristic axis are merged to obtain the signal instruction integrated descriptor for each characteristic axis.

5. The monitoring signal visualization method for implantable medical monitoring sensors as described in any one of claims 1 to 4, characterized in that, The process of integrating the signal command descriptor and the signal descriptor to identify one or more target signal sampling points in the monitored signal segment includes: Based on the characteristic axis of the signal descriptor, the signal descriptor is arranged in order, and the signal descriptor with the target characteristic axis is selected from the signal descriptor according to the order arrangement result to obtain the current signal descriptor; The signal command integration descriptor corresponding to the target feature axis is selected from the signal command integration descriptors to obtain the current signal command integration descriptor. Based on the current signal instruction integrated descriptor and the current signal descriptor, a target signal descriptor is extracted from the monitored signal segment, and one or more target signal sampling points in the monitored signal segment are determined according to the target signal descriptor; The step of extracting the target signal descriptor from the monitored signal segment based on the current signal instruction integrated descriptor and the current signal descriptor includes: Based on the current signal instruction integrated descriptor and the current signal descriptor, the preset target signal descriptor is optimized, and the optimized target signal descriptor is determined as the preset target signal descriptor; The process jumps to the step of selecting the target feature axis signal descriptor from the signal descriptors based on the ordered arrangement result and repeats until each signal descriptor is the current signal descriptor, thus obtaining the target signal descriptor.

6. The monitoring signal visualization method for implantable medical monitoring sensors as described in claim 5, characterized in that, The step of optimizing the preset target signal descriptor based on the current signal instruction and the current signal descriptor includes: Based on the current signal instruction integrated descriptor and the current signal descriptor, the preset target signal descriptor is optimized to obtain the basic target signal descriptor, and the basic target signal descriptor is determined as the preset target signal descriptor; The process of integrating the current signal instruction and the current signal descriptor to optimize the preset target signal descriptor is repeated until the preset number of optimizations is met, and the optimized target signal descriptor is obtained.

7. The monitoring signal visualization method for implantable medical monitoring sensors as described in claim 6, characterized in that, The preset target signal descriptor includes one or more sub-target signal descriptors. The step of optimizing the preset target signal descriptor based on the current signal instruction integration descriptor and the current signal descriptor to obtain a basic target signal descriptor includes: Adaptive weight allocation is performed on the sub-target signal descriptors to obtain candidate target signal descriptors; Cross-modal weight allocation is performed on the candidate target signal descriptor, the current signal command integrated descriptor, and the current signal descriptor to obtain the current target signal descriptor; The current target signal descriptor is projected onto a preset target representation domain to obtain the basic target signal descriptor.

8. The monitoring signal visualization method for implantable medical monitoring sensors as described in any one of claims 1 to 4, characterized in that, The implicit representation of the monitored signal segment to obtain a signal descriptor with one or more characteristic axes includes: The monitored signal segment is implicitly represented at multiple levels to obtain one or more basic signal descriptors at feature levels; The basic signal descriptor is normalized to obtain one or more candidate signal descriptors with characteristic axes; Based on a preset feature fusion path, the candidate signal descriptors are fused to obtain signal descriptors with one or more feature axes.

9. A computer system, characterized in that, include: One or more processors; Memory; One or more computer programs; The one or more computer programs are stored in the memory and configured to be executed by the one or more processors, and when the one or more computer programs are executed by the processors, they implement the method as described in any one of claims 1 to 8.

Images