In-vivo monitoring information processing method and system based on implantable medical control device

Abstract

The application provides a kind of based on implantable medical control instrument living body monitoring information processing method and system, it is related to data processing field, based on the abnormal data distribution result of reference living body monitoring information to the abnormal data array of reference living body monitoring information Distance calculation is carried out to the array representation of monitoring information of to-be-processed living body monitoring information, similarity score is obtained, the distribution position inference mark of abnormal data is obtained by the distribution position inference mark of abnormal data, and the preset abnormal data recognition model is adjusted to obtain the convergence abnormal data recognition model.The distribution position inference mark of abnormal data in to-be-processed living body monitoring information is obtained by distance calculation, so as to obtain the distribution position of a small amount of to-be-processed living body monitoring information based on a small amount of expert annotated reference living body monitoring information inference mark, a large number of prior knowledge samples can be obtained only with a small amount of prior annotation samples, and sample augmentation is completed.Model training.In addition, the precision of the convergence abnormal data recognition model for marking the distribution position of abnormal data can be improved.

Description

Technical Field

[0001] This invention relates to the field of data processing, and more specifically, to a method and system for processing in vivo monitoring information based on implantable medical control devices. Background Technology

[0002] In modern medicine, especially in the area of ​​in vivo monitoring involving implantable medical devices, the accurate identification and localization of abnormal data in the monitoring information is crucial. With the continuous development of medical technology, implantable medical devices can collect a large amount of in vivo monitoring information, such as continuous monitoring data of human physiological parameters (e.g., heart rate, blood pressure, blood sugar). Traditional methods often rely on a large number of expert-annotated samples for model training. This means that a significant amount of manpower, resources, and time are required to obtain sufficient annotation samples. For example, when performing anomaly analysis on long-term monitoring data of cardiac function, experts need to meticulously annotate each potentially abnormal data point and its distribution. This process not only requires experts with high levels of expertise but also involves a huge workload, making the acquisition of sufficient annotation samples a challenging task. Due to the difficulty in obtaining sufficient annotation samples, the model may not be able to fully learn the various feature patterns in the data, resulting in low accuracy when identifying the location of abnormal data distributions in new in vivo monitoring information. Simultaneously, limited samples also restrict the efficiency of model training, making it difficult for the model to quickly converge to an ideal state, thus affecting the overall efficiency of the anomaly identification system. Furthermore, traditional processing methods may lack effective methods for sample augmentation. When dealing with complex live surveillance information, if the number and diversity of samples cannot be effectively increased, the model may overfit to a limited number of samples, resulting in poor generalization ability in practical applications and an inability to accurately identify the distribution location of abnormal data under different conditions. Summary of the Invention

[0003] The purpose of this invention is to provide a method and system for processing in vivo monitoring information based on implantable medical control devices. This invention is achieved as follows:

[0004] In a first aspect, the present invention provides a method for processing in vivo monitoring information based on an implantable medical control device, the method comprising:

[0005] A reference liveness monitoring information database and a liveness monitoring information database to be processed are obtained. The reference liveness monitoring information database includes one or more reference liveness monitoring information and abnormal data distribution results of the reference liveness monitoring information. The liveness monitoring information database to be processed includes multiple liveness monitoring information to be processed. The liveness monitoring information in both the reference liveness monitoring information database and the liveness monitoring information database to be processed is obtained based on data collected by implantable medical control devices.

[0006] Based on the abnormal data distribution results, the distance between the abnormal data array representation of the reference liveness monitoring information and the monitoring information array representation of the liveness monitoring information to be processed is calculated to obtain the similarity score of the liveness monitoring information to be processed;

[0007] Based on the similarity score, inference is performed to obtain the distribution location inference markers of abnormal data in the liveness monitoring information to be processed;

[0008] Based on the distribution location inference markers, a first prior distribution heatmap of the distribution location of abnormal data in the liveness monitoring information to be processed is obtained;

[0009] Based on the first prior distribution heat map and the live monitoring information to be processed, the preset abnormal data identification model is adjusted to obtain a converged abnormal data identification model, wherein the converged abnormal data identification model is used to identify and mark the distribution location of abnormal data in the live monitoring information to be processed.

[0010] Optionally, the step of reasoning based on the similarity score to obtain the distribution location inference markers of abnormal data in the liveness monitoring information to be processed includes:

[0011] Based on the similarity score, inference of abnormal data distribution information is performed to obtain the first abnormal data distribution information probability density of the liveness monitoring information to be processed;

[0012] Step scan the probability density of the first abnormal data distribution information to obtain the regional peak response intensity of each first scan region of the probability density of the first abnormal data distribution information.

[0013] If the peak response intensity of the first scanned region is greater than a first set threshold, then the first distribution information corresponding to the peak response intensity of the first scanned region is determined as the distribution location inference marker.

[0014] Optionally, the method further includes:

[0015] If the peak response intensity of the first scanned area is not greater than the first set threshold, then the liveness monitoring information to be processed will be used as the liveness monitoring information to be processed.

[0016] Based on the convergent anomaly data identification model, reasoning is performed on the liveness monitoring information to be processed to obtain the marking results of the distribution location of the anomaly data in the liveness monitoring information to be processed.

[0017] Optionally, the step of reasoning about the liveness monitoring information to be processed based on the convergent anomaly data identification model to obtain the labeling results of the distribution location of anomaly data in the liveness monitoring information to be processed includes:

[0018] Based on the convergent anomaly data identification model, reasoning is performed on the liveness monitoring information to be processed to obtain the probability density of the second anomaly data distribution information of the liveness monitoring information to be processed.

[0019] A step scan is performed on the probability density of the second abnormal data distribution information to determine the regional peak response intensity of each second scan region of the probability density of the second abnormal data distribution information.

[0020] If the peak response intensity of the second scanning area is greater than the second set threshold, then the second distribution information corresponding to the peak response intensity of the second scanning area is determined as the marking result of the abnormal data distribution location in the liveness monitoring information to be processed.

[0021] Optionally, the step of reasoning about the liveness monitoring information to be processed based on the convergent anomaly data identification model to obtain the labeling results of the distribution location of anomaly data in the liveness monitoring information to be processed includes:

[0022] Based on the convergent anomaly data identification model, reasoning is performed on the liveness monitoring information to be processed to obtain the probability density of the second anomaly data distribution information of the liveness monitoring information to be processed.

[0023] Based on the probability density of the second abnormal data distribution information, the distribution information with the highest probability of abnormal data distribution information is obtained, and the result is determined as the marking result of the abnormal data distribution location in the live monitoring information to be processed.

[0024] Optionally, the reference liveness monitoring information database includes first reference liveness monitoring information and second reference liveness monitoring information. The step of calculating the distance between the abnormal data array representation of the reference liveness monitoring information and the monitoring information array representation of the liveness monitoring information to be processed, based on the abnormal data distribution results, to obtain a similarity score for the liveness monitoring information to be processed, includes:

[0025] Based on the abnormal data distribution results of the first reference liveness monitoring information, an abnormal data array representation of the first reference liveness monitoring information is obtained; based on the abnormal data distribution results of the second reference liveness monitoring information, an abnormal data array representation of the second reference liveness monitoring information is obtained;

[0026] Based on the live organism monitoring information to be processed, a characterization array is extracted to obtain a monitoring information array representation of the live organism monitoring information to be processed;

[0027] Distance calculation is performed between the abnormal data array representation of the first reference liveness monitoring information and the monitoring information array representation to obtain a first similarity metric value between the first reference liveness monitoring information and the liveness monitoring information to be processed;

[0028] Distance calculation is performed between the abnormal data array representation of the second reference liveness monitoring information and the monitoring information array representation to obtain a second similarity metric value between the second reference liveness monitoring information and the liveness monitoring information to be processed;

[0029] The first similarity metric and the second similarity metric are combined to obtain the similarity score.

[0030] Optionally, the step of adjusting the preset abnormal data identification model to obtain a converged abnormal data identification model based on the first prior distribution heatmap and the liveness monitoring information to be processed includes:

[0031] Based on a preset abnormal data identification model, reasoning is performed on the live monitoring information to be processed to obtain a first reasoning distribution heat map of the distribution location of abnormal data in the live monitoring information to be processed.

[0032] Based on the first inference distribution heatmap and the first prior distribution heatmap, the first error result of the preset abnormal data identification model is obtained;

[0033] Based on the first error result, the preset abnormal data identification model is adjusted to obtain a converged abnormal data identification model.

[0034] Optionally, the step of reasoning about the liveness monitoring information to be processed based on a preset abnormal data identification model to obtain a first inference distribution heatmap of the distribution location of abnormal data in the liveness monitoring information to be processed includes:

[0035] Based on a preset abnormal data identification model, the characterization array of the live monitoring information to be processed is extracted to obtain the initial array representation of the live monitoring information to be processed.

[0036] Based on a preset abnormal data identification model, the initial array representation of the liveness monitoring information to be processed is subjected to a convolution operation to obtain the convolution array representation of the liveness monitoring information to be processed.

[0037] A residual connection operation is performed based on the initial array representation and the convolution array representation to obtain the first inference distribution heatmap of the liveness monitoring information to be processed.

[0038] Optionally, the step of adjusting the preset abnormal data identification model based on the first error result to obtain a converged abnormal data identification model includes:

[0039] Based on the abnormal data distribution results of the reference liveness monitoring information, a second prior distribution heatmap of the abnormal data distribution location in the reference liveness monitoring information is obtained;

[0040] Based on the preset abnormal data identification model, the reference liveness monitoring information is inferred to obtain a second inference distribution heat map of the abnormal data distribution location in the reference liveness monitoring information;

[0041] Based on the second inference distribution heatmap and the second prior distribution heatmap, the second error result of the preset abnormal data identification model is obtained;

[0042] Based on the first error result and the second error result, the preset abnormal data identification model is adjusted to obtain a converged abnormal data identification model.

[0043] Optionally, obtaining a first prior heatmap of the distribution locations of abnormal data in the liveness monitoring information to be processed, based on the distribution location inference markers, includes:

[0044] Based on the distribution location inference marker, the liveness monitoring information to be processed is binary encoded to obtain the encoded monitoring information of the liveness monitoring information to be processed;

[0045] The encoded monitoring information is smoothed and filtered to obtain a first prior heatmap of the distribution location of abnormal data in the liveness monitoring information to be processed.

[0046] In a second aspect, the present invention provides a computer system comprising: 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, they implement the method described above.

[0047] The beneficial effects of this invention are as follows: Based on the abnormal data distribution results of reference liveness monitoring information, this invention calculates the distance between the abnormal data array representation of the reference liveness monitoring information and the monitoring information array representation of the liveness monitoring information to be processed, obtaining a similarity score for the liveness monitoring information to be processed. This score is used to infer the distribution location inference markers of the abnormal data in the liveness monitoring information to be processed. Thus, a converged abnormal data identification model is obtained by adjusting the preset abnormal data identification model using these distribution location inference markers. Firstly, because calculating the distance between the abnormal data array representation of the reference liveness monitoring information and the monitoring information array representation of the liveness monitoring information to be processed allows for the inference of the distribution location inference markers of the abnormal data in the liveness monitoring information to be processed, enabling the acquisition of the distribution location of a small amount of the liveness monitoring information to be processed based on the inference markers of the reference liveness monitoring information with a small amount of expert annotations. This allows for the acquisition of a large number of prior knowledge samples using only a small number of pre-annotated samples, thus enhancing the sample augmentation for model training. Furthermore, because vector distance calculation can augment samples and generate labels for a large amount of liveness monitoring information to be processed, the pre-set anomaly identification model can be tuned using this liveness monitoring information, improving the accuracy of the converged anomaly identification model in marking the distribution location of anomalies. In summary, this invention can reduce costs and improve efficiency and accuracy in liveness monitoring information processing. Attached Figure Description

[0048] Figure 1 This is a flowchart of a method for processing in vivo monitoring information based on an implantable medical control device, provided by an embodiment of the present invention.

[0049] Figure 2 This is a schematic diagram of the composition of a computer system provided in an embodiment of the present invention. Detailed Implementation

[0050] In this embodiment of the invention, the execution entity of the live monitoring information processing method based on implantable medical control devices is a computer system, including but not limited to servers, personal computers, laptops, tablets, and smartphones. Figure 1 As shown, the method includes:

[0051] Step S100: Obtain a reference liveness monitoring information database and a liveness monitoring information database to be processed. The reference liveness monitoring information database includes one or more reference liveness monitoring information entries and abnormal data distribution results of the reference liveness monitoring information entries. The liveness monitoring information database to be processed includes multiple liveness monitoring information entries to be processed. The liveness monitoring information entries in both the reference liveness monitoring information database and the liveness monitoring information database to be processed are obtained based on data collected from implantable medical control devices.

[0052] In step S100, the reference living body monitoring information database contains one or more reference living body monitoring information. This information is acquired by the computer system from a specific data source, which is related to the implantable medical device. For example, in a cardiac disease monitoring scenario, the implantable medical device may be a monitoring chip attached to a pacemaker. The reference living body monitoring information can be various data about cardiac function, such as heart rate, systolic and diastolic blood pressure, and electrocardiogram (ECG) signals. Taking heart rate as an example, it may be a record of the number of heartbeats per minute over a period of time (such as 24 hours), forming a time series data, such as [70, 72, 75, 73, …]. For ECG signals, it may be the amplitude values ​​of electrical signals collected every millisecond, forming a complex waveform data sequence. This data reflects the physiological characteristics of a living body in a normal or specific state and is an important basis for subsequent abnormal data judgment and processing.

[0053] The reference liveness monitoring database also contains the distribution results of abnormal data within the reference liveness monitoring information, i.e., the location of abnormal data within the reference liveness monitoring information. For example, continuing with cardiac monitoring, if abnormal fluctuations in heart rate are found within the aforementioned heart rate data over certain time periods, assuming that the heart rate suddenly rises to 120-130 beats per minute between the 10th and 15th minute, while the normal range should be 60-100 beats per minute, then the computer system will mark this time period as the location of the abnormal data distribution. For electrocardiogram (ECG) signals, if the amplitude or shape of certain waveforms is abnormal, such as sharp peaks or irregular waveforms lasting for a certain period, the computer system will accurately record the starting position and duration of this abnormal waveform within the entire ECG signal sequence; these records constitute the abnormal data distribution results.

[0054] In more complex physiological monitoring scenarios, such as monitoring brain neural activity, implantable medical modulators might be electrodes implanted in the brain to monitor neuronal firing activity. The reference in vivo monitoring information includes data such as the frequency and intensity of these neuronal firings. If, within a certain time period, the firing frequency of neurons in a specific brain region suddenly increases or decreases abnormally, the computer system records the location of this anomaly throughout the monitoring period, as the distribution of abnormal data. This is of great significance for studying brain diseases or abnormalities in nervous system function.

[0055] The pending liveness monitoring database contains multiple pending liveness monitoring data. This information is also collected by implantable medical devices. For example, in the scenario of blood glucose monitoring for diabetic patients, an implantable blood glucose monitor continuously collects the patient's blood glucose levels. These blood glucose levels change constantly with the patient's diet, exercise, medication, and other factors. The pending liveness monitoring information acquired by the computer system is a sequence of these blood glucose values, such as [5.0, 6.5, 4.8, 7.2, …], where each value represents the patient's blood glucose concentration at a specific moment. Similarly, in renal function monitoring, implantable medical devices may monitor the levels of various components in urine, such as creatinine and urea nitrogen; the pending liveness monitoring information is a data sequence of these component levels changing over time.

[0056] In muscle activity monitoring, if an implantable medical device is a sensor capable of monitoring muscle electrical signals, then the in vivo monitoring information to be processed includes data such as the intensity and frequency of electrical signals generated by the muscles under different activity states (e.g., rest, exercise, exertion). This data can help doctors understand the functional state of the muscles and whether there are any potential muscle diseases or injuries.

[0057] The computer system acquires these live animal monitoring information databases to be processed in order to compare and analyze them with reference live animal monitoring information, thereby identifying the possible distribution of abnormal data and performing corresponding processing, such as the identification, marking and analysis of abnormal data. This is an important foundation of the entire live animal monitoring information processing method.

[0058] Step S200: Based on the abnormal data distribution results, calculate the distance between the abnormal data array representation of the reference live monitoring information and the monitoring information array representation of the live monitoring information to be processed, and obtain the similarity score of the live monitoring information to be processed.

[0059] In this embodiment, step S200 calculates the distance between the abnormal data array representation of the reference liveness monitoring information and the monitoring information array representation of the liveness monitoring information to be processed, based on the abnormal data distribution results in the reference liveness monitoring information database obtained in step S100, thereby obtaining a similarity score for the liveness monitoring information to be processed. The reference liveness monitoring information is collected based on implantable medical control devices, and this information contains rich physiological data. Taking cardiac monitoring as an example, assuming that the reference liveness monitoring information collected by the implantable device includes data such as heart rate, blood pressure, and electrocardiogram signals. Among these data, there are some data that have been marked as abnormal, such as abnormal values ​​of heart rate. If the heart rate values ​​over a period of time are regarded as a sequence, such as [70, 72, 150, 73, 75] (where 150 is an abnormal value), this sequence can be converted into an array representation that is easy for computers to process. This array may contain not only the raw numerical values, but also timestamps or other auxiliary information associated with those values, such as [(70,t1), (72, t2), (150, t3), (73, t4), (75, t5)], where ti represents the corresponding time point. This is one possible form of representing anomaly data arrays based on liveness monitoring information.

[0060] Similarly, for the array representation of the monitoring information to be processed, the computer system needs to perform similar processing on the monitoring information itself. Continuing with cardiac monitoring as an example, if the monitoring information to be processed contains newly acquired heart rate data, such as [71, 73, 74, 72, 76], the computer system will convert it into a monitoring information array representation, possibly [(71, t1'), (73, t2'), (74, t3'), (72, t4'), (76, t5')]. This array representation accurately reflects the characteristics of the monitoring information to be processed and has the same structure as the abnormal data array representation of the reference monitoring information, facilitating subsequent distance calculations.

[0061] Distance calculations performed by computer systems are based on the two array representations mentioned above. There are various methods for distance calculation, such as Euclidean distance calculation. Taking a simple two-dimensional array as an example, suppose we have an anomalous data array A=[(1, 2), (3, 4)] representing reference liveness monitoring information and a monitoring information array B=[(2, 3), (4, 5)] representing the liveness monitoring information to be processed. The Euclidean distance formula is... For the first element (1, 2) in array A and the first element (2, 3) in array B, calculate the distance as follows: The distance between the second element (3, 4) in array A and the second element (4, 5) in array B is... Then, these distances are combined (such as summation, averaging, etc.) to obtain a value representing the distance between the two.

[0062] In more complex liveness monitoring scenarios, such as when considering multi-dimensional data like heart rate, blood pressure, and ECG signals simultaneously, the dimensionality of the array increases. Assume the abnormal data array of the reference liveness monitoring information is represented as A=[(70,120, Waveform 1), (72, 118, Waveform 2), (150, 130, Waveform 3)] (where 70 is heart rate, 120 is blood pressure, and Waveform 1 is the ECG signal waveform, etc.), and the monitoring information array of the liveness monitoring information to be processed is represented as B=[(71, 119, Waveform 4), (73, 121, Waveform 5), (74, 120, Waveform 6)]. The computer system will use distance calculation methods suitable for high-dimensional data, such as Manhattan distance or cosine similarity. Taking cosine similarity as an example, it measures the cosine of the angle between two vectors. For high-dimensional vectors A and B, by calculating their cosine similarity, a value between -1 and 1 can be obtained, representing the degree of similarity between the two. The closer this value is to 1, the more similar the information represented by the two arrays; the closer it is to -1, the less similar they are; and the closer it is to 0, the less linear the relationship between them.

[0063] Through this distance calculation, the computer system obtains a distance metric between each piece of liveness monitoring information to be processed and a reference piece of liveness monitoring information. Then, based on this distance metric, the computer system further calculates a similarity score for the liveness monitoring information to be processed. The similarity score is an indicator that comprehensively reflects the degree of similarity between the liveness monitoring information to be processed and the reference liveness monitoring information. For example, the smaller the distance metric, the higher the similarity score may be. The computer system may use a mapping function to convert the distance metric into a similarity score. Assuming the distance metric d ranges from [0, +∞), the similarity score s can be calculated using the function s = 1 / (1+d) (this is just a simple example function). Thus, when d = 0, s = 1, indicating perfect similarity; as d approaches +∞, s approaches 0, indicating complete dissimilarity.

[0064] This similarity score provides crucial input for subsequent steps in the entire liveness monitoring information processing flow, such as the reasoning based on the similarity score in step S300. By calculating the similarity score, the computer system can establish a quantitative relationship between the reference liveness monitoring information and the liveness monitoring information to be processed. This allows it to infer and predict the distribution of abnormal data in the unknown liveness monitoring information to be processed based on the known distribution of abnormal data in the reference liveness monitoring information. This helps improve the accuracy and efficiency of medical monitoring, especially when processing large amounts of liveness monitoring information, enabling the rapid screening of potentially abnormal information and providing strong support for further diagnosis and treatment.

[0065] Step S300: Based on the similarity score, reasoning is performed to obtain the distribution location inference markers of abnormal data in the live monitoring information to be processed.

[0066] In this embodiment, the computer system performs inference based on the similarity score obtained in step S200 to obtain inference markers for the distribution locations of abnormal data in the liveness monitoring information to be processed. The similarity score reflects the degree of similarity between the liveness monitoring information to be processed and the reference liveness monitoring information. The reference liveness monitoring information contains known distributions of abnormal data, so the possible distribution locations of abnormal data in the liveness monitoring information to be processed can be inferred through the similarity score. For example, in liveness monitoring of blood glucose levels, the reference liveness monitoring information is case data from multiple cases with typical blood glucose fluctuations (including abnormal fluctuations). These data have been processed to obtain abnormal data distribution results, and a similarity score was calculated between them and the new patient blood glucose monitoring information to be processed in step S200. The computer system can use a classifier (such as softmax) to perform inference based on the similarity score.

[0067] Step S400: Based on the distribution location inference markers, obtain the first prior distribution heat map of the distribution location of abnormal data in the live monitoring information to be processed.

[0068] In this embodiment, the computer system obtains a distribution location inference marker for the abnormal data in the liveness monitoring information to be processed in step S300. This inference marker is derived based on similarity scoring and represents a preliminary judgment on the possible distribution location of the abnormal data. In step S400, the computer system uses this distribution location inference marker to obtain a first prior heatmap of the distribution location of the abnormal data in the liveness monitoring information to be processed.

[0069] Taking cardiac monitoring as an example, suppose that in the previous steps, the computer system obtained a distribution location inference label by processing a series of reference in vivo monitoring information (including heart rate, electrocardiogram signals, etc.) and the in vivo monitoring information to be processed. This label may indicate that in the in vivo monitoring information to be processed, abnormal heartbeats are likely to occur within a specific time period. For example, within a two-hour monitoring period, the interval between the 30th and 40th minutes is the most likely location for abnormal heartbeat data. This label is a preliminary judgment based on the similarity between data and an understanding of overall physiological patterns.

[0070] The computer system uses this distributed location inference marker to perform binary encoding on the liveness monitoring information to be processed. Continuing with the cardiac monitoring example, assume the entire monitoring period is divided into several small time units, such as one minute per unit. For each time unit, if it falls within the potentially abnormal interval indicated by the distributed location inference marker (minutes 30-40), it is encoded as 1, indicating that an abnormality may exist in this unit; otherwise, it is encoded as 0. In this way, the entire liveness monitoring information to be processed is converted into a sequence of encoded monitoring information composed of 0s and 1s. For example, for a 60-minute monitoring record, the encoded sequence might be [0, 0, 0, …, 0, 1, 1, 1, …, 1, 0, 0, 0, …, 0], where the position corresponding to minute 30-40 is 1.

[0071] The significance of this binary encoding lies in transforming the inference markers into a form that a computer system can further process to generate a priori distribution heatmaps. It simplifies complex inferences about the location of outlier data distributions into a binary state representation, highlighting areas where anomalies may exist.

[0072] Next, the computer system performs smoothing filtering on the coded monitoring information to obtain a first prior distribution heatmap. Smoothing filtering is used to make the resulting prior distribution heatmap more consistent with the actual physiological data changes, reducing the discreteness and noise that may be introduced by binary coding. For example, the computer system may use a weighted average method during smoothing filtering. For each time unit coded as 1, it not only affects its own heat value but also influences its adjacent time units according to a certain weight. For example, in the cardiac monitoring example above, for the 30th minute coded as 1, it will influence the heat values ​​of the 29th and 31st minutes with a weight of 0.5, and the heat values ​​of the 28th and 32nd minutes with a weight of 0.3, and so on.

[0073] Through this smoothing filtering process, the computer system obtains the first prior distribution heatmap. This heatmap can be viewed as a probability distribution, representing the prior probability distribution of abnormal data occurring at different locations in the liveness monitoring information to be processed. In the heatmap, areas with higher values ​​indicate a greater likelihood of abnormal data occurring at that location. For example, in the case of heart monitoring, the 30-40 minute interval might appear as a high-value area in the heatmap, while other intervals might appear as low-value areas, indicating a higher prior probability of abnormal heartbeat data occurring in those intervals.

[0074] Let's take the monitoring of specific substance concentrations in blood as an example. Assume that in previous steps, the distribution location inference markers indicated a possible anomaly in the concentration of a specific substance during a certain intermediate period in the blood collection process. The computer system performs binary encoding on the blood monitoring data corresponding to this period and then applies smoothing filtering. If there is a certain continuity between different stages during blood collection, such as during the absorption, metabolism, and excretion of substances, where the concentrations of substances in adjacent periods are correlated, then smoothing filtering can better reflect the impact of this continuity on the distribution of abnormal data. The resulting first prior distribution heatmap can accurately show the prior probability distribution of abnormal concentrations of specific substances at different times throughout the entire blood collection process. In the entire live animal monitoring information processing flow, the first prior distribution heatmap provides crucial input for subsequent steps, such as step S500, where the preset abnormal data identification model is adjusted based on the heatmap and the live animal monitoring information to be processed. This heatmap is not merely a simple result presentation; it further integrates the inference and judgment in previous steps into a quantifiable and intuitive analytical form, helping to improve the accuracy and effectiveness of the entire live animal monitoring information processing method. In this way, computer systems can delve deeper into the distribution patterns of abnormal data in live animal monitoring information, providing stronger support for medical diagnosis and control.

[0075] Step S500: Based on the first prior distribution heat map and the live monitoring information to be processed, the preset abnormal data identification model is adjusted to obtain a converged abnormal data identification model. The converged abnormal data identification model is used to identify and mark the distribution location of abnormal data in the live monitoring information to be processed.

[0076] In step S500 of this embodiment, the computer system adjusts the preset abnormal data identification model based on the first prior distribution heatmap and the liveness monitoring information to obtain a converged abnormal data identification model. The preset abnormal data identification model is a model pre-built by the computer system to identify abnormal data in the liveness monitoring information. It may be based on various machine learning algorithms, such as neural network algorithms. Taking a simple neural network model as an example, it includes an input layer, a hidden layer, and an output layer. The input layer receives liveness monitoring information, which may be pre-processed values, such as standardized values ​​of heart rate, blood pressure, and electrocardiogram signals in cardiac monitoring. The hidden layer performs complex nonlinear transformations on the input data, while the output layer outputs the judgment result regarding the abnormal data, such as outputting a probability value indicating the existence of abnormal data or directly marking the location of the abnormal data.

[0077] In step S500, the computer system infers the location of abnormal data in the live monitoring information to be processed based on a preset abnormal data identification model, obtaining a first inference distribution heatmap. This process involves feature extraction and analysis of the input live monitoring information. Taking blood glucose monitoring as an example, assume that the live monitoring information received by the preset abnormal data identification model is a sequence of blood glucose values ​​measured at regular time intervals (e.g., 15 minutes) over a period of time (e.g., one day). The model first extracts features from these blood glucose values, which may include calculating statistical features such as the average, variance, and fluctuation range of the blood glucose values, and may also include analyzing the trend of blood glucose values ​​over time, such as whether the blood glucose values ​​are rising, falling, or remaining stable. Then, the model infers based on these features to generate a first inference distribution heatmap. This heatmap represents the model's judgment on the probability of abnormal blood glucose data occurring at each time point or time period. For example, after breakfast, blood glucose values ​​usually rise due to the digestion and absorption of food, but if the rise exceeds the normal range, the model may display a higher value at the corresponding heatmap location for this time period, indicating that there is a higher probability of abnormally high blood glucose data occurring during this time period.

[0078] Next, the computer system obtains the first error result of the preset anomaly identification model based on the first inference distribution heatmap and the first prior distribution heatmap. The first prior distribution heatmap is obtained in step S400 based on the distribution location inference markers, reflecting the prior probability distribution of the anomaly data distribution locations obtained based on the inference in the previous steps. The first inference distribution heatmap is the result obtained by the preset anomaly identification model based on its own algorithm and structure to infer the data. The computer system determines the model error by comparing these two heatmaps. For example, the mean squared error (MSE) method can be used to calculate the error. Assuming that the heat value at a certain time point t in the first prior distribution heatmap is h1, and the heat value at the corresponding time point t in the first inference distribution heatmap is h2, then the mean squared error is calculated as follows: , where n is the number of time points in the heatmap. This error result reflects the degree of difference between the current inference result of the preset anomaly identification model and the prior distribution heatmap obtained based on the previous steps.

[0079] Then, the computer system uses this first error result to fine-tune the preset anomaly identification model to obtain a converged anomaly identification model. The fine-tuning process aims to make the model's output closer to the actual distribution of anomaly data. In neural network models, fine-tuning typically involves adjusting the model's weight parameters. For example, if a large error is found in a certain region when calculating the first error result, it indicates that the model's judgment of anomaly data in that region is inaccurate. The computer system will use the backpropagation algorithm to propagate the error from the output layer to the input layer, adjusting the weights of the intermediate hidden layers and the connections between the input and hidden layers. Continuing with the blood glucose monitoring example, if the model is found to have a large error in judging post-lunch blood glucose fluctuations, then during the fine-tuning process, the model will adjust the weight parameters related to processing post-lunch blood glucose data. If a neuron's weight in judging the upward trend of post-lunch blood glucose is too high, leading to an excessively high probability of predicting abnormal blood glucose levels, then this weight value will be reduced during the fine-tuning process.

[0080] Let's illustrate the entire process of step S500 with another example. In in vivo monitoring of kidney function, a pre-defined abnormal data identification model receives data such as urine composition, blood creatinine, and blood urea nitrogen levels collected from implantable medical devices as in vivo monitoring information to be processed. The model first extracts and analyzes features from this complex physiological data to generate a first inference distribution heatmap regarding the location of abnormal kidney function data. This heatmap may show that the kidney's excretory function may be abnormal during certain time periods, such as the kidney's reabsorption of certain substances at night, which would show a higher probability of abnormality on the heatmap for the corresponding nighttime period. Then, the computer system compares this first inference distribution heatmap with a first prior distribution heatmap obtained based on the previous steps (this heatmap may be based on similarity analysis of a large number of reference data from cases with normal and abnormal kidney function and the in vivo monitoring information to be processed) to calculate a first error result. If a significant discrepancy is found between the first inference heatmap's assessment of nocturnal renal reabsorption function abnormalities and the first prior heatmap—for example, the first inference heatmap showing an excessively high probability of abnormality while the first prior heatmap shows a low probability—the computer system will adjust the weight parameters in the preset anomaly identification model based on this error. This might involve adjusting the weights of input features related to nocturnal renal function, such as reducing the weight of a feature related to changes in nocturnal urine composition, as this feature might be overemphasized by the model, leading to errors. Through multiple such adjustments, the preset anomaly identification model gradually converges, ultimately resulting in a converged anomaly identification model. This converged model can more accurately identify the distribution locations of abnormal data in the in vivo monitoring information being processed, thereby improving the accuracy and reliability of the entire in vivo monitoring information processing method.

[0081] As one implementation method, the reference liveness monitoring information database includes first reference liveness monitoring information and second reference liveness monitoring information. Step S200 involves calculating the distance between the abnormal data array representation of the reference liveness monitoring information and the monitoring information array representation of the liveness monitoring information to be processed, based on the abnormal data distribution results, to obtain a similarity score for the liveness monitoring information to be processed. This may include:

[0082] Step S210: Obtain the abnormal data array representation of the first reference liveness monitoring information based on the abnormal data distribution results of the first reference liveness monitoring information; obtain the abnormal data array representation of the second reference liveness monitoring information based on the abnormal data distribution results of the second reference liveness monitoring information;

[0083] Step S220: Extract the characterization array based on the live monitoring information to be processed to obtain the monitoring information array representation of the live monitoring information to be processed;

[0084] Step S230: Perform distance calculation on the abnormal data array representation and the monitoring information array representation of the first reference live body monitoring information to obtain the first similarity metric value between the first reference live body monitoring information and the live body monitoring information to be processed;

[0085] Step S240: Calculate the distance between the abnormal data array representation and the monitoring information array representation of the second reference liveness monitoring information to obtain a second similarity metric between the second reference liveness monitoring information and the liveness monitoring information to be processed;

[0086] Step S250: Integrate the first similarity measure and the second similarity measure to obtain a similarity score.

[0087] In this embodiment of the application, the specific implementation of step S200 includes steps S210-S250. This series of steps is executed by a computer system, which aims to obtain the similarity score of the live monitoring information to be processed by performing a series of operations on the abnormal data array representation of the reference live monitoring information and the monitoring information array representation of the live monitoring information to be processed, based on the abnormal data distribution results in the reference live monitoring information database.

[0088] In step S210, the computer system first operates on the first reference in vivo monitoring information. In a medical monitoring scenario, taking cardiovascular system monitoring as an example, assume the first reference in vivo monitoring information includes data from an implantable cardiac monitoring device. This data covers multiple aspects such as heart rate, blood pressure, and electrocardiogram (ECG) signals. Heart rate data may be measurements taken over 24 hours at certain time intervals (e.g., recorded every 5 minutes), blood pressure data includes systolic and diastolic blood pressure, and ECG signals are more complex continuous waveform data.

[0089] This reference in vivo monitoring information already contains known abnormal data distributions. For example, through annotation by professional medical personnel or preliminary data analysis, it has been determined that there were abnormal fluctuations in heart rate during specific time periods (such as 10:00-11:00 AM), blood pressure values ​​exceeding the normal range between 3:00-4:00 PM, and irregular waveforms in electrocardiogram signals at a certain time during the night.

[0090] The computer system constructs an array representation of abnormal data based on the distribution of these abnormal data. For abnormal heart rate data, assuming the heart rate values ​​between 10:00 and 11:00 AM are [105, 110, 108, 112, 106] (the normal range is assumed to be 60-100 beats / minute), these abnormal heart rate values ​​are combined with the corresponding timestamps to form array elements, such as [(105, 10:00), (110, 10:05), (108, 10:10), (112, 10:15), (106, 10:20)]. For abnormal blood pressure data, the systolic and diastolic blood pressure values ​​between 3 PM and 4 PM are [(150, 90), (148, 88), (152, 92)] (the normal systolic blood pressure range is assumed to be 90-139 mmHg, and the diastolic blood pressure range is 60-89 mmHg). Combined with the time stamp, this can be represented as [((150, 90), 15:00), ((148, 88), 15:05), ((152, 92), 15:10)]. Due to the complexity of electrocardiogram signals, abnormal data may be represented by extracting specific feature values ​​(such as peak values ​​and intervals of the waveform). Assuming that the feature value extracted during a certain period at night is [5, 3, 4] (these values ​​are quantized features), the corresponding array elements could be [(5, 0:30), (3, 0:35), (4, 0:40)] (0:30-0:40 is the abnormal period at night). By combining these abnormal data array elements of heart rate, blood pressure, and electrocardiogram signals in a specific order, an abnormal data array representation of the first reference in vivo monitoring information is formed. Similarly, the computer system performs a similar operation for the second reference in vivo monitoring information. Taking nervous system monitoring as an example, suppose an implantable medical control device monitors data such as the firing frequency and neurotransmitter concentration of brain neurons as the second reference in vivo monitoring information.

[0091] The known distribution of anomalous data indicates that during specific cognitive tasks (such as memory recall tasks), there are abnormal fluctuations in neuronal firing frequency and deviations from normal ranges in neurotransmitter concentrations in specific brain regions. For example, during a memory recall task, the anomalous neuronal firing frequency within 3-5 minutes is [150, 160, 155, 148, 152] (the normal range is assumed to be 100-130 times / second), forming an array element [(150, 3:00), (160, 3:05), (155, 3:10), (148, 3:15), (152, 3:20)] with the corresponding task time. Simultaneously, during the same task, the abnormal values ​​for neurotransmitter concentration (hypothetically dopamine concentration) in a certain brain region were [0.8, 0.9, 0.75] (the normal range is assumed to be 0.5-0.7), which, combined with the time stamp, resulted in [(0.8, 3:00), (0.9, 3:05), (0.75, 3:10)]. Combining the abnormal data array elements of neuronal firing frequency and neurotransmitter concentration yields the abnormal data array representation of the second reference in vivo monitoring information.

[0092] In step S220, the computer system extracts the representation array based on the live monitoring information to be processed, and obtains the monitoring information array representation of the live monitoring information to be processed.

[0093] Taking liver function monitoring as an example, suppose the in vivo monitoring information to be processed is data on the changes of various liver function-related indicators (such as alanine aminotransferase, aspartate aminotransferase, bilirubin, etc.) in the blood over time, collected by an implantable liver monitoring device.

[0094] For alanine aminotransferase (ALT), assuming measurements are taken every 2 hours over 12 hours, the resulting data would be [40, 45, 50, 55, 60, 65] (the normal range is assumed to be 0-40 U / L). The computer system extracts a characterization array from this data, which may not only involve simply recording the numerical values ​​but also incorporating information such as the measurement time, for example, [(40, 0:00), (45, 2:00), (50, 4:00), (55, 6:00), (60, 8:00), (65, 10:00)]. For aspartate aminotransferase (AST), the measured values ​​are [35, 40, 42, 48, 50, 52] (normal range assumed to be 0-40 U / L), and the extracted characterization array may be [(35, 0:00), (40, 2:00), (42, 4:00), (48, 6:00), (50, 8:00), (52, 10:00)]. The measured values ​​for bilirubin are [1.2, 1.5, 1.8, 2.0, 2.2, 2.5] (normal range assumed to be 0-1.0 mg / dL), and the extracted array is represented as [(1.2, 0:00), (1.5, 2:00), (1.8, 4:00), (2.0, 6:00), (2.2, 8:00), (2.5, 10:00)].

[0095] By combining these liver function-related indicator representation arrays in a certain order, we obtain the monitoring information array representation of the in vivo monitoring information to be processed.

[0096] In step S230, the computer system performs distance calculation on the abnormal data array representation and the monitoring information array representation of the first reference live body monitoring information to obtain a first similarity metric value between the first reference live body monitoring information and the live body monitoring information to be processed.

[0097] There are various methods for distance calculation; here, we will use Euclidean distance as an example. Assume that an element in the abnormal data array representation of the first reference live body monitoring information is abnormal heart rate data [(105, 10:00), (110, 10:05), (108, 10:10), (112, 10:15), (106, 10:20)], and the heart rate data in the monitoring information array representation of the live body monitoring information to be processed is [(90, 10:00), (95, 10:05), (92, 10:10), (98, 10:15), (96, 10:20)].

[0098] For each corresponding heart rate value at a given time point, calculate the Euclidean distance. For example, at the time point 10:00, the distance is... At 10:05, the distance And so on. Then these distance values ​​are processed together, for example, by averaging them to obtain the first similarity metric. Suppose that after calculating the distance and averaging the heart rate values ​​at all corresponding time points, the first similarity metric obtained is 12 (this is just an example value).

[0099] If we consider an array representation of multiple physiological indicators (such as heart rate, blood pressure, and electrocardiogram signals), the calculation process becomes more complex. The distance is calculated for each array element of each indicator, and then these distances are combined according to certain weights (which may be pre-set based on the importance of the indicators) to obtain the first similarity metric.

[0100] In step S240, the computer system performs distance calculation on the abnormal data array representation and the monitoring information array representation of the second reference liveness monitoring information to obtain a second similarity metric between the second reference liveness monitoring information and the liveness monitoring information to be processed.

[0101] Taking the previously mentioned neurological monitoring and liver function monitoring as examples. Suppose that the neuronal firing frequency in the abnormal data array representation of the second reference in vivo monitoring information is [(150, 3:00), (160, 3:05), (155, 3:10), (148, 3:15), (152, 3:20)], and the data of a certain neurological-related indicator (assumed to be neural reaction time) in the monitoring information array representation of the in vivo monitoring information to be processed is [(130, 3:00), (140, 3:05), (135, 3:10), (138, 3:15), (142, 3:20)].

[0102] Using the same Euclidean distance calculation method, at the time 3:00, the distance... At 3:05, the distance The distance at each time point is calculated sequentially and then processed comprehensively.

[0103] For comparisons between neurotransmitter concentrations and liver function indicators, similar distance calculations can be performed if a transformation method is used to map them to a comparable space (e.g., through normalization). Assume that after calculation and comprehensive processing, the resulting second similarity metric is 15 (again, an example value).

[0104] In step S250, the computer system integrates the first similarity measure and the second similarity measure to obtain a similarity score. Various integration methods can be used, such as a weighted average. Assuming the first similarity measure is 12 and the second similarity measure is 15, and based on the importance of the first and second reference liveness monitoring information (this importance may be determined based on medical experience or prior data analysis), the weight of the first reference liveness monitoring information is set to 0.4, and the weight of the second reference liveness monitoring information is set to 0.6. Then the similarity score is... .

[0105] Another integration method could be based on a combination of some nonlinear functions. For example, suppose... , where a and b are parameters predetermined based on data characteristics and processing requirements.

[0106] Through this integration method, the computer system obtains a similarity score that comprehensively reflects the degree of similarity between the liveness monitoring information to be processed and the reference liveness monitoring information (including the first reference liveness monitoring information and the second reference liveness monitoring information). This score will serve as an important basis for subsequent steps (such as the reasoning operation in step S300).

[0107] As one implementation method, step S300, which involves reasoning based on similarity scores to obtain distribution location inference markers for abnormal data in the liveness monitoring information to be processed, may include:

[0108] Step S310: Based on the similarity score, perform abnormal data distribution information inference to obtain the probability density of the first abnormal data distribution information of the liveness monitoring information to be processed;

[0109] Step S320: Perform a step scan on the probability density of the first abnormal data distribution information to obtain the regional peak response intensity of each first scan region of the probability density of the first abnormal data distribution information;

[0110] Step S330: If the peak response intensity of the first scan area is greater than the first set threshold, then the first distribution information corresponding to the peak response intensity of the first scan area is determined as the distribution location inference marker.

[0111] In step S310, abnormal data distribution information is inferred based on the similarity score to obtain the first abnormal data distribution information probability density of the liveness monitoring information to be processed. The similarity score is a value obtained in step S200 by performing complex distance calculations and integration on the reference liveness monitoring information and the liveness monitoring information to be processed. It reflects the degree of similarity between the liveness monitoring information to be processed and the reference liveness monitoring information.

[0112] For example, the process of inferring the probability density of the first abnormal data distribution information based on the similarity score to obtain the liveness monitoring information to be processed includes, for example:

[0113] The computer system first normalizes the obtained similarity scores. Assume there are n similarity scores. Calculate the normalized value for each similarity score. For example, if there are three similarity scores Then their normalized values ​​are respectively .

[0114] Then, the weight vector is determined based on the normalized similarity score. ,in These weights will be used to construct the probability density function, representing the relative importance of each similarity score in inferring information about the distribution of anomalous data.

[0115] Next, the computer system selects a suitable probability density function form, such as a Gaussian Mixture Model (GMM). A GMM is a probability model composed of multiple Gaussian distributions, suitable for modeling data with complex distributions. Suppose k Gaussian distributions are chosen to construct the GMM, with parameters for each distribution... ,in It is the mean. It is the standard deviation. .

[0116] The parameters for each Gaussian distribution can be determined based on data features from reference liveness monitoring information and the weights of similarity scores. For example, for the mean... The similarity score can be determined based on the statistical characteristics of the data in the reference liveness monitoring information that have a high similarity score (i.e., a larger weight) with the liveness monitoring information to be processed. If the reference liveness monitoring information contains physiological data from different time periods, and the data from a certain time period has a high similarity score with the liveness monitoring information to be processed, then the mean of the data from that time period may have a higher similarity score. It has made a significant contribution to the standard deviation. Similarly, it can be determined based on the dispersion of the data and the weight of the similarity score.

[0117] Next, based on the determined Gaussian distribution parameters and weight vector, a mixture probability density function is constructed. , where x represents the possible distribution location of the abnormal data. This probability density function represents the probability distribution of abnormal data at different locations in the liveness monitoring information to be processed, i.e., the probability density of the first abnormal data distribution information.

[0118] Considering the prior distribution of the reference liveness monitoring data, the computer system obtains the prior distribution information of abnormal data from the reference liveness monitoring data. For example, if the reference liveness monitoring data is about blood glucose levels, and the probability distribution of abnormal increases or decreases in blood glucose under certain specific conditions (such as a specific time period after a meal or a specific exercise state) is known, this prior distribution information can be represented as a probability density function. .

[0119] The probability density function constructed based on similarity scores The prior distribution probability density function of the reference liveness monitoring information The data is then fused. A simple fusion method is weighted summation to obtain the final probability density of the first anomaly data distribution. α is a weighting coefficient ranging from [0,1], used to balance the importance of inference results based on similarity scores and prior distribution information. For example, when α = 0.8, it indicates a greater preference for inference results based on similarity scores, but also considers a certain proportion of prior distribution information.

[0120] Taking cardiovascular system monitoring as an example, suppose the reference in vivo monitoring information includes multi-dimensional data such as heart rate, blood pressure, and electrocardiogram signals. This data exhibits a specific pattern over a period of time, and the distribution of its abnormal data is known. The in vivo monitoring information to be processed is also similar data about the cardiovascular system. If the similarity score is high, it indicates that the in vivo monitoring information to be processed is quite similar to the reference in vivo monitoring information in terms of data patterns. The computer system will use this similarity score, combined with the abnormal data distribution pattern in the reference in vivo monitoring information, to infer the distribution information of abnormal data in the in vivo monitoring information to be processed. For heart rate, assuming it occurs within 30 minutes after waking up in the morning, based on the similarity score and reference information, the computer system infers that the probability density function of abnormal data occurrences may follow a normal distribution, such as... , where t represents time (within these 30 minutes). This could be the most likely time point (e.g., 10 minutes) to occur, determined based on reference information and similarity scores. It represents the standard deviation of the probability density distribution (determined based on data characteristics and similarity).

[0121] In this way, the computer system constructs the probability density of the first abnormal data distribution information of different physiological indicators in the live monitoring information to be processed, based on the similarity score and by comprehensively considering the abnormal data distribution patterns in the reference live monitoring information. These probability density functions comprehensively describe the possible distribution of abnormal data in the live monitoring information to be processed.

[0122] In step S320, after obtaining the probability density of the first abnormal data distribution information, the computer system performs a step-by-step scan. Taking the cardiovascular system monitoring example above, for the abnormal data probability density function of heart rate... Assuming a scanning step of 1 minute, within 30 minutes of waking up in the morning, starting from the first minute, the computer system calculates the probability density value corresponding to each time point t. .

[0123] When t = 1 minute, calculate When t=2 minutes, calculate And so on, until t=30 minutes.

[0124] For the probability density function of blood pressure Similarly, scans were performed at 1-minute intervals within 1-2 hours (60-120 minutes) after lunch to calculate the probability density value at each time point.

[0125] In the example of respiratory system monitoring, the probability density function for respiratory rate... The probability density function of blood oxygen saturation within 10-20 minutes Within 5-10 minutes, scan according to the set step size (e.g., 30 seconds per step) and calculate the probability density value corresponding to each step size.

[0126] During the step scan, the computer system determines the peak response intensity of each first scan region. Taking heart rate as an example, suppose the 30 minutes after waking up in the morning are divided into three regions: 0-10 minutes is the first region, 11-20 minutes is the second region, and 21-30 minutes is the third region.

[0127] Within the first region, the computer system finds the maximum probability density during the scanning process. The intensity corresponding to this maximum value is the regional peak response intensity of the first region. Assuming that within this region, at t=5 minutes... Reaching the maximum value .

[0128] For the probability density function of blood pressure, different regions are divided within 1-2 hours after lunch, such as 30-60 minutes as one region, 61-90 minutes as another region, and 91-120 minutes as a third region. Similarly, the maximum value of the probability density is found in each region and determined as the regional peak response intensity of that region.

[0129] In the respiratory system, the probability density functions of respiratory rate and blood oxygen saturation are used to determine the regional peak response intensity within their respective scanning regions in the same way. For example, a respiratory rate of 10-13 minutes is considered a region, and the maximum probability density value within this region is used as the regional peak response intensity; a blood oxygen saturation of 5-7 minutes is considered a region, and the regional peak response intensity within this region is determined.

[0130] In step S330, the computer system compares the peak response intensity of the first scan region with a first preset threshold. Assuming this is in cardiovascular system monitoring, the peak response intensity of the first region for heart rate... The first set threshold value is .

[0131] if Then the computer system will determine the first distribution information corresponding to the peak response intensity of the first region as the distribution location inference marker. For example, if If the corresponding time is t=5 minutes, then the heart rate-related information at this 5-minute mark (such as the heart rate value and its relationship with surrounding heart rates) is identified as a distribution location inference marker. This means that there may be anomalous data distributions near this time point. The same comparison operation is performed for physiological indicators such as blood pressure, respiratory rate, and blood oxygen saturation. If the peak response intensity of a certain region is greater than the corresponding first set threshold, then the distribution information (such as time and numerical range) corresponding to the peak response intensity of that region is identified as a distribution location inference marker.

[0132] In this way, in steps S310-S330, the computer system gradually constructs the probability density of abnormal data distribution information based on similarity scores, performs step-by-step scanning to determine the peak response intensity of the region, and determines the distribution location inference marker based on comparison with the set threshold value, providing a preliminary judgment basis for the distribution location of abnormal data in the live monitoring information to be processed for subsequent processing steps.

[0133] As one implementation method, the method may further include:

[0134] Step S301: If the peak response intensity of the first scan area is not greater than the first set threshold, then the liveness monitoring information to be processed is used as the liveness monitoring information to be processed.

[0135] Step S302: Based on the convergent anomaly data identification model, reason about the liveness monitoring information to be processed to obtain the labeling results of the distribution location of the anomaly data in the liveness monitoring information to be processed.

[0136] In this embodiment of the application, a derivative implementation of step S300 includes steps S301-S302. This series of steps is executed by a computer system and is a supplementary operation for inferring and marking the distribution location of abnormal data under specific conditions.

[0137] In step S301, the peak response intensity of the first scanned region is first compared with a first set threshold value. This comparison is based on the peak response intensity of the region obtained in the previous step (such as step S320) and the preset threshold value.

[0138] Taking renal function monitoring as an example, suppose the in vivo monitoring information to be processed includes data such as creatinine, blood urea nitrogen concentration, and glomerular filtration rate in the blood. In the previous steps, the computer system inferred the distribution information of abnormal data based on similarity scores, obtaining the first probability density of abnormal data distribution information for these data. For example, for creatinine concentration, probability density functions were constructed for different time periods of the day, and scans were performed at certain step sizes to determine the regional peak response intensity of each scan area.

[0139] Assuming the first set threshold is established based on extensive clinical experience and normal physiological data ranges, for scan regions with creatinine concentrations in the morning (e.g., 0:00-6:00), the regional peak response intensity is... The first set threshold value is .if This indicates that although there is a certain possibility of abnormal data distribution within this scanned area, according to the established criteria, this possibility is not sufficient to be identified as a high-risk area.

[0140] Similarly, comparisons were made within their respective scanning regions for data such as blood urea nitrogen concentration and glomerular filtration rate. For example, the regional peak response intensity of blood urea nitrogen concentration during the afternoon period (12:00-18:00) was compared. With the corresponding first set threshold value Comparison, if ; Regional peak response intensity of glomerular filtration rate during the evening period (18:00-24:00) With the corresponding first set threshold value Comparison, if .

[0141] When the peak response intensity of all the first scanned regions is not greater than the first set threshold, the computer system treats the liveness monitoring information to be processed as liveness monitoring information to be processed. This means that further processing is needed to determine the distribution location of the abnormal data. Because in the previous steps, although preliminary inference and scanning were performed through similarity scoring, no regions that clearly meet the conditions (regional peak response intensity greater than the threshold) were found. Therefore, the distribution location of the abnormal data cannot be directly determined for inference labeling. Additional operations are required, such as processing based on the convergent abnormal data identification model (step S302).

[0142] Continuing with the example of kidney function monitoring, if the peak response intensity of creatinine, blood urea nitrogen concentration, and glomerular filtration rate in their respective scanning areas is not greater than the corresponding threshold value, the computer system considers that the current inference and scanning results alone cannot accurately determine the distribution location of the abnormal data. Therefore, it retains the complete in vivo monitoring information to be processed for further analysis based on a convergent abnormal data identification model. This is because there may be complex, yet unidentified, abnormal data distribution patterns that require a more powerful model to uncover.

[0143] In step S302, the computer system infers the location of abnormal data distribution in the liveness monitoring information to be processed based on the convergent anomaly data identification model to obtain the labeling results. The convergent anomaly data identification model is obtained through a series of adjustments (such as in step S500), which enables a more in-depth and accurate analysis of the liveness monitoring information.

[0144] Taking nervous system monitoring as an example, suppose the in vivo monitoring information to be processed includes data such as the firing frequency of brain neurons, neurotransmitter concentration, and electroencephalogram (EEG) signals. After receiving this data, the convergent abnormal data identification model first extracts and analyzes its features.

[0145] Regarding neuronal firing frequency, the model may analyze characteristics such as average firing frequency, frequency fluctuation amplitude, and firing frequency periodicity in different brain regions. For example, in the frontal lobe region of the brain, the normal neuronal firing frequency is between 10-30 times / second, with small fluctuation amplitude and a certain periodicity. If the neuronal firing frequency in the frontal lobe region of the in vivo monitoring data shows an average firing frequency of 40 times / second, with large fluctuation amplitude and periodic disorder, the model will record these characteristics as abnormalities. Regarding neurotransmitter concentrations, such as dopamine and serotonin, the model will analyze whether their concentrations are within the normal range and their ratios with other neurotransmitter concentrations. Assuming the normal dopamine to serotonin concentration ratio is 1:2, if this ratio changes to 1:3 in the in vivo monitoring data, the model will identify this change as a possible manifestation of abnormal data.

[0146] For EEG signals, the model extracts features such as frequency, amplitude, and phase. For example, normal alpha EEG frequencies are between 8-13 Hz, with relatively stable amplitudes. If the alpha EEG frequency in the liveness monitoring data being processed exceeds this range or the amplitude fluctuates significantly, the model uses these features as the basis for analyzing abnormal data. Through comprehensive analysis of this data, the convergent abnormal data identification model constructs a probability density function or other representation of the distribution location of abnormal data in the liveness monitoring data being processed. For example, it constructs a matrix of the probability of abnormal data distribution in different brain regions, where each element represents the probability of abnormal data appearing in the corresponding region.

[0147] In this process, the model may be based on a large amount of training data and knowledge accumulated during the tuning process. For example, it may have learned specific neuronal firing frequencies, neurotransmitter concentrations, and EEG signal change patterns in certain disease states (such as Parkinson's disease, which is associated with abnormal neuronal activity in the substantia nigra region of the brain). This allows it to match and infer from the features in the in vivo monitoring information to determine the possible distribution location of abnormal data.

[0148] In this way, the computer system performs comprehensive and in-depth reasoning on the live monitoring information to be processed based on the convergent anomaly data identification model, providing a basis for the final determination of the labeling results of the distribution location of the anomaly data.

[0149] As one implementation method, step S302, which involves reasoning based on a convergent anomaly data identification model to obtain a labeling result of the distribution location of anomaly data in the to-be-processed liveness monitoring information, may include:

[0150] Step S3021: Based on the convergent anomaly data identification model, reason about the liveness monitoring information to be processed to obtain the probability density of the second anomaly data distribution information of the liveness monitoring information to be processed.

[0151] Step S3022: Perform a step scan on the probability density of the second abnormal data distribution information to determine the regional peak response intensity of each second scan region of the probability density of the second abnormal data distribution information;

[0152] Step S3023: If the peak response intensity of the second scanning area is greater than the second set threshold, then the second distribution information corresponding to the peak response intensity of the second scanning area is determined as the marking result of the abnormal data distribution location in the live monitoring information to be processed.

[0153] In step S3021, reasoning is performed on the liveness monitoring information to be processed based on the converged anomaly identification model to obtain the probability density of the second anomaly data distribution information of the liveness monitoring information to be processed. The converged anomaly identification model is obtained through the previous steps (such as the calibration process in step S500), and it already has the ability to analyze the anomaly data distribution in the liveness monitoring information.

[0154] Taking cardiovascular system monitoring as an example, suppose the in vivo monitoring information to be processed includes data such as heart rate, blood pressure, and electrocardiogram signals. After receiving this data, the convergence anomaly identification model will process it according to its internal algorithms and parameters.

[0155] For heart rate data, the model may consider long-term trends, short-term fluctuations, and differences from normal physiological patterns. For example, the resting heart rate of a normal adult is typically between 60 and 100 beats per minute. The model will analyze whether the heart rate in the in vivo monitoring data falls within this range and whether its fluctuation pattern conforms to normal physiological rhythms. If the heart rate is consistently higher than 100 beats per minute or lower than 60 beats per minute, and this does not conform to normal physiological fluctuations (such as fluctuations caused by exercise or emotional excitement), the model will use these data characteristics as the basis for constructing the probability density of the second abnormal data distribution information.

[0156] Suppose that the heart rate data over a period of time (e.g., 24 hours) is [70, 72, 80, 90, 105, 110, 100, 95, 90, 85, 80, 75, 70, 65, 60, 55, 50, 55, 60, 65, 70, 75, 80, 85] (unit: beats / minute). The model will identify heart rates in the range of 105-110 beats / minute as potentially abnormal. Based on the overall distribution of heart rate, the degree of deviation from the normal range, and similar patterns in historical data, the model may construct a probability density function for the distribution of abnormal heart rate data, for example, using a Gaussian Mixture Model (GMM). Let... , where x represents the heart rate and k is the number of Gaussian distributions (assuming k=2). These are the weights of the i-th Gaussian distribution. and These are the mean and variance of the i-th Gaussian distribution, respectively. In this example, a Gaussian distribution might describe the normal range of heart rate (...). Another Gaussian distribution describes the range of abnormal heart rate ( ), weight This indicates a higher probability of a normal heart rate, but there is still a possibility of an abnormal heart rate.

[0157] For blood pressure data, systolic and diastolic blood pressure are analyzed separately. The normal systolic blood pressure range is assumed to be 90-139 mmHg, and the diastolic blood pressure range is assumed to be 60-89 mmHg. If the blood pressure data in the in vivo monitoring information to be processed is [(120,80), (130, 85), (140, 90), (150, 95), (145, 92), (135, 88), (130, 85), (125,82), (120, 80)] (unit: mmHg), the model will identify a systolic blood pressure of 140-150 mmHg as potentially abnormal. A similar method is used to construct the probability density function for the distribution of abnormal blood pressure data, such as... Where y represents the blood pressure value (which can be either systolic or diastolic pressure), and m is the number of Gaussian distributions. and These are the weights, mean, and variance of the i-th Gaussian distribution, respectively.

[0158] Due to the complexity of electrocardiogram (ECG) signals, models may extract features such as QRS complex width, ST segment shift, and T wave morphology. For example, a normal QRS complex width is between 0.06 and 0.10 seconds. If the QRS complex width in the ECG signal being processed is 0.12 seconds, the model will incorporate this feature into the probability density function for constructing the second abnormal data distribution information. Through comprehensive analysis of these ECG signal features, a probability density function for the abnormal data distribution of ECG signals is constructed. Its construction method can also be based on Gaussian mixture models or other suitable probability models.

[0159] By analyzing data such as heart rate, blood pressure, and electrocardiogram signals, the computer system constructs a second probability density of abnormal data distribution information of the live monitoring information to be processed based on a convergent abnormal data identification model. This probability density comprehensively describes the probability of abnormal data distribution under different data characteristics.

[0160] In step S3022, after obtaining the probability density of the second abnormal data distribution information, the computer system needs to perform a step-by-step scan. Continuing with the cardiovascular system monitoring example above, for the probability density of the second abnormal data distribution information of heart rate... Assuming a scan interval of 5 scans per minute. If When x = 50 times / minute, calculate The value; when x = 55 times / minute, calculate again. The value is calculated, and so on, scanning across the entire range of possible heart rate values. For example, starting from 40 beats / minute and scanning up to 120 beats / minute, the probability density value corresponding to each step size is calculated.

[0161] The probability density of the second abnormal data distribution information for blood pressure Assuming a scanning step of 5 mmHg. If For systolic blood pressure, the scan starts from 80 mmHg and goes up to 160 mmHg, and the probability density value corresponding to each step is calculated; for diastolic blood pressure, the scan starts from 50 mmHg and goes up to 100 mmHg, and the probability density value corresponding to each step is calculated.

[0162] The probability density of the second abnormal data distribution information of electrocardiogram signals Due to the complexity of ECG signal characteristics, the definition of step size may be adjusted according to specific characteristics. For example, for QRS complex width, a scan is performed with a step size of 0.01 seconds, starting from 0.05 seconds and scanning up to 0.15 seconds, and the probability density value corresponding to each step size is calculated; for ST segment deviation, a scan is performed with a step size of 0.1 mV, and the probability density value corresponding to each step size is calculated within a certain voltage range.

[0163] During the step-scan process, the computer system determines the peak response intensity of each second scan region. Taking heart rate as an example, suppose 40-60 beats / minute is divided into the first region, 61-80 beats / minute into the second region, 81-100 beats / minute into the third region, and 101-120 beats / minute into the fourth region.

[0164] Within the first region, the computer system finds the maximum probability density during the scanning process. The intensity corresponding to this maximum value is the regional peak response intensity of the first region. For example, in this region, when x = 50 times / minute, P_{hf}(50) reaches its maximum value. .

[0165] For blood pressure data, assuming that for systolic blood pressure, the first region is defined as 80-90 mmHg, the second as 91-110 mmHg, the third as 111-130 mmHg, and the fourth as 131-160 mmHg. The maximum probability density value within each region is identified as the regional peak response intensity (GPR) for that region. For example, in the second region, when y = 100 mmHg, P_{bp}(100) reaches its maximum value; this maximum value is the GPR for that region.

[0166] For electrocardiogram (ECG) signals, taking the QRS complex width as an example, the intervals 0.05-0.07 seconds are divided into five regions: 0.071-0.09 seconds, 0.091-0.11 seconds, 0.111-0.13 seconds, and 0.131-0.15 seconds. The maximum probability density value within each region is then identified as the region's peak response intensity.

[0167] In step S3023, the computer system compares the peak response intensity of the second scan region with a second preset threshold. Assuming that in cardiovascular system monitoring, the peak response intensity of the fourth region (101-120 beats / minute) of the heart rate is... The second set threshold value is .

[0168] if > Then, the computer system determines the second distribution information corresponding to the peak response intensity of the fourth region as the labeling result of the abnormal data distribution location in the liveness monitoring information to be processed. For example, if The corresponding heart rate is 110 beats / minute. Therefore, the relevant information about the heart rate at this 110 beats / minute (such as the relationship with the previous and next heart rates, the time of occurrence, etc.) is identified as the labeling result. This means that there may be abnormal data distribution in the vicinity of this heart rate.

[0169] For blood pressure data, if the peak response intensity of a certain region (such as the fourth region of systolic blood pressure, 131-160 mmHg) is greater than the corresponding second set threshold, then the distribution information corresponding to the peak response intensity of that region (such as blood pressure value, time of occurrence, etc.) is determined as the marking result of the abnormal data distribution location in the live monitoring information to be processed.

[0170] For electrocardiogram (ECG) signals, the peak response intensity of each region is also compared with the corresponding second set threshold. For example, the peak response intensity of the third region (0.091-0.11 seconds) of the QRS complex width is compared. ,if > Then, the QRS complex width information corresponding to the peak response intensity in this region (such as the specific width value, the relationship with other ECG signal characteristics, etc.) is determined as the labeling result, indicating that there may be abnormal data distribution within this QRS complex width range.

[0171] In this way, in steps S3021-S3023, the computer system, based on the convergent abnormal data identification model, performs reasoning, scanning and comparison on the live monitoring information to be processed, and finally determines the labeling results of the distribution location of abnormal data in the live monitoring information to be processed, providing an important basis for further medical diagnosis and decision-making.

[0172] As another implementation of step S302, step S302, based on the convergent anomaly data identification model, infers the location of the abnormal data distribution in the liveness monitoring information to be processed, and obtains the marking results of the distribution of abnormal data in the liveness monitoring information to be processed, which may include:

[0173] Step S302A: Based on the convergent anomaly data identification model, reason about the liveness monitoring information to be processed to obtain the probability density of the second anomaly data distribution information of the liveness monitoring information to be processed;

[0174] Step S302B: Based on the probability density of the second abnormal data distribution information, obtain the distribution information with the highest probability of abnormal data distribution information, and determine it as the marking result of the abnormal data distribution location in the live monitoring information to be processed.

[0175] In this embodiment of the application, another implementation of step S302 includes steps S302A-S302B. This series of steps is executed by a computer system and is intended to process the live monitoring information to be processed based on a convergent abnormal data identification model, thereby obtaining the marking results of the distribution location of abnormal data in the live monitoring information to be processed.

[0176] In step S302A, reasoning is performed on the liveness monitoring information to be processed based on the converged anomaly identification model to obtain the probability density of the second anomaly distribution information of the liveness monitoring information to be processed. The converged anomaly identification model is obtained through the previous calibration process (such as the calibration operation in step S500), and it has been optimized to accurately identify the anomaly distribution in the liveness monitoring information.

[0177] Taking liver function monitoring as an example, suppose the in vivo monitoring information to be processed includes data on indicators such as alanine aminotransferase (ALT), aspartate aminotransferase (AST), and bilirubin in the blood. After receiving this data, the convergence anomaly identification model will analyze it based on its internal algorithm and pre-learned knowledge.

[0178] For example, for alanine aminotransferase (ALT) data, the normal range is assumed to be 0-40 U / L. If the ALT values ​​in the in vivo monitoring data to be processed are [30, 35, 45, 50, 55, 60, 58, 52, 48, 42] (unit: U / L), the model will consider the degree of deviation of these values ​​from the normal range and the trend of these values. For example, a gradual increase from 30-35 U / L to 60 U / L might indicate an anomaly. The model might use some probability distribution model to describe the probability of such anomalies, assuming a log-normal distribution to construct the probability density function. , where x represents the value of ALT. Based on the characteristics of the data, determine the parameters of the log-normal distribution, such as the mean. and standard deviation For example, by learning from a large amount of normal and abnormal liver function data, the model determines that for ALT data, =35, =10, thus constructing .

[0179] For aspartate aminotransferase (AST) data, the normal range is assumed to be 0-40 U / L. Assume the AST values ​​in the in vivo monitoring data to be processed are [32, 38, 42, 48, 50, 52, 55, 50, 45, 40] (unit: U / L). The model will also analyze its deviation from the normal range and its trend. Since AST and ALT have a certain correlation in liver function, the model may consider their relationship comprehensively. For example, if ALT is elevated while AST is also elevated, this synergistic change may increase the likelihood of abnormality. The model may construct a joint probability density function. , where x represents the value of ALT and y represents the value of AST. This joint probability density function can be based on a multivariate normal distribution or other distribution models suitable for describing the relationship between the two.

[0180] For bilirubin data, assuming a normal range of 0-1.0 mg / dL, the bilirubin values ​​in the in vivo monitoring data to be processed are [0.8, 0.9, 1.1, 1.2, 1.3, 1.5, 1.4, 1.3, 1.2, 1.1] (unit: mg / dL). The model considers the elevation of bilirubin and its interaction with other liver indicators. For example, elevated bilirubin may occur simultaneously with elevated ALT and AST, which may indicate a problem with the liver's bile excretion function. The model may construct a probability density function for bilirubin. , where z represents the bilirubin value. Assume a gamma distribution is used. Parameters are determined through data analysis. and ,like =2, =0.5.

[0181] By analyzing data such as alanine aminotransferase (ALT), aspartate aminotransferase (AST), and bilirubin, the computer system constructs a probability density of the second abnormal data distribution information of the in vivo monitoring information to be processed based on a convergent abnormal data identification model. This probability density comprehensively describes the probability distribution of abnormal data under different liver function indicators.

[0182] In step S302B, after obtaining the probability density of the second abnormal data distribution information, the computer system will use this probability density to obtain the distribution information with the highest probability of abnormal data distribution information. Continuing with the liver function monitoring example above, for the constructed probability density function of alanine aminotransferase (ALT)... The computer system will find the value that makes ALT equal to the value that is possible within the entire range of possible ALT values. Find the x-value that represents the maximum value.

[0183] Suppose that calculations show P_{ALT}(x) reaches its maximum value when x = 55 U / L. This x = 55 U / L is the value with the highest probability of outlier distribution information in the ALT data.

[0184] The joint probability density function for aspartate aminotransferase (AST) and alanine aminotransferase (ALT) The computer system will find the value in the two-dimensional space (x represents the value of ALT, y represents the value of AST) that makes Find the (x, y) combination that yields the maximum value. Suppose calculations show that when x = 52 U / L (ALT value) and y = 50 U / L (AST value), The maximum value is obtained. This (x,y) combination represents the distribution information with the highest probability of outlier distribution information when considering the relationship between ALT and AST.

[0185] For the probability density function of bilirubin Suppose that calculations show that when z = 1.3 mg / dL, The maximum value is obtained. This z=1.3 mg / dL is the value with the highest probability of abnormal data distribution information in the bilirubin data.

[0186] The computer system identifies the distribution information with the highest probability of abnormal data as the labeling result for the abnormal data distribution location in the in vivo monitoring information to be processed. For liver function monitoring, this labeling result includes the values ​​of various liver function indicators (ALT, AST, bilirubin, etc.) when abnormalities are most likely to occur.

[0187] For example, the labeling results can be expressed as: ALT=55 U / L, AST=50 U / L, bilirubin=1.3 mg / dL. This labeling result provides medical personnel with crucial information about the distribution of abnormal data in the in vivo monitoring data to be processed, helping them to further analyze potential problems with liver function, such as whether there is hepatocellular damage, bile excretion disorders, etc.

[0188] In this way, in steps S302A-S302B, the computer system, based on the convergent anomaly identification model, infers the distribution information of the live monitoring information to be processed and obtains the distribution information with the highest probability of anomaly distribution information. Finally, it determines the labeling results of the distribution location of anomaly data in the live monitoring information to be processed, providing important basis for the diagnosis and subsequent treatment of liver diseases. This implementation method, when processing live monitoring information, can effectively utilize the capabilities of the convergent anomaly identification model to accurately locate the distribution location of anomaly data, improving the accuracy and reliability of the entire medical monitoring system.

[0189] As one implementation method, step S400, obtaining a first prior heatmap of the distribution locations of abnormal data in the liveness monitoring information to be processed based on the distribution location inference markers, may include:

[0190] Step S410: Based on the distribution location inference marker, perform binary encoding on the live monitoring information to be processed to obtain the encoded monitoring information of the live monitoring information to be processed;

[0191] Step S420: Perform smoothing filtering on the coded monitoring information to obtain the first prior heatmap of the distribution location of abnormal data in the live monitoring information to be processed.

[0192] In this embodiment of the application, the specific implementation of step S400 includes steps S410-S420. This series of steps is executed by a computer system and aims to obtain a first prior heatmap of the distribution location of abnormal data in the live monitoring information to be processed based on the distribution location inference marker.

[0193] In step S410, the liveness monitoring information to be processed is binary encoded based on the distribution location inference markers obtained in step S300. The distribution location inference markers are an inference of the possible distribution locations of abnormal data, which provides the basis for binary encoding.

[0194] Taking kidney function monitoring as an example, suppose the in vivo monitoring information to be processed includes data such as creatinine, blood urea nitrogen concentration and glomerular filtration rate in the blood. These data are collected at certain time intervals (such as every hour) over a period of time (e.g., 24 hours).

[0195] Assume that the distribution location inference markers indicate a possible anomalous distribution of creatinine concentration data during the 9-11 AM time period. The computer system will perform binary coding on the creatinine concentration data for the entire 24-hour period. For each time point within the 9-11 AM time period, the creatinine concentration data will be encoded as 1, indicating a possible anomaly; while for creatinine concentration data in other time periods (such as 0-8 AM, 12-12 AM), the encoding will be 0, indicating that anomalies are unlikely to exist in these time periods based on the inference markers.

[0196] For example, the creatinine concentration data sequence is [C1, C2, C3, …, C9, C10, C11, …, C24] (where Ci represents the creatinine concentration value corresponding to each hour). If 9:00 AM corresponds to C9, 10:00 AM corresponds to C10, and 11:00 AM corresponds to C11, then the encoded sequence may be [0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0].

[0197] Through this binary encoding method, the computer system transforms the complex liveness monitoring information to be processed into a simple binary state representation based on the distribution location inference label, highlighting areas where anomalies may exist, and laying the foundation for the subsequent generation of prior distribution heat maps.

[0198] In step S420, after completing the binary encoding, the computer system performs smoothing filtering on the obtained encoded monitoring information to obtain a first prior heatmap of the distribution locations of abnormal data in the liveness monitoring information to be processed. The purpose of smoothing filtering is to reduce the discreteness caused by binary encoding and make the obtained heatmap more consistent with the actual distribution pattern of abnormal data.

[0199] Taking renal function monitoring as an example, for the coded monitoring information of creatinine concentration [0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], the computer system uses a weighted average smoothing filter method. Assuming a simple 3-point weighted average filter is used, for each point in the coded sequence, its value will be updated to the weighted average of itself and its immediate and next-to-next points. For example, for point 9 (coded as 1), its new value will be calculated as... (Assuming the weight of adjacent points is 0.2 and the weight of the point itself is 0.6), we get 0.8. For the 10th point (coded as 1), the new value is calculated as... For point 11 (coded as 1), the new value is .

[0200] After this smoothing and filtering process, these updated coded values ​​constitute the first prior distribution heatmap. In this heatmap, regions with higher values ​​indicate a greater probability of anomalous data occurring at that location. For example, for creatinine concentration, the heatmap value is relatively high for the time period of 9-11 AM, indicating a higher prior probability of anomalous creatinine concentration data during this time period; for urea nitrogen concentration, the heatmap value is relatively high for the time period of 3-5 PM; and for glomerular filtration rate, the heatmap value is relatively high for the time period of 7-9 PM.

[0201] This first prior distribution heatmap provides important input for subsequent steps (such as the calibration of the preset abnormal data identification model in step S500). It intuitively represents the prior probability distribution of the abnormal data distribution location, which helps to improve the accuracy and effectiveness of the entire live monitoring information processing method.

[0202] As one implementation method, step S500, based on the first prior distribution heatmap and the liveness monitoring information to be processed, adjusts the preset abnormal data identification model to obtain a converged abnormal data identification model, which may include:

[0203] Step S510: Based on the preset abnormal data identification model, reason about the live monitoring information to be processed to obtain the first reasoning distribution heat map of the distribution location of abnormal data in the live monitoring information to be processed.

[0204] Step S520: Based on the first inference distribution heatmap and the first prior distribution heatmap, obtain the first error result of the preset abnormal data identification model;

[0205] Step S530: Adjust the preset abnormal data identification model based on the first error result to obtain a converged abnormal data identification model.

[0206] In this embodiment of the application, the specific implementation of step S500 includes steps S510-S530. This series of steps is executed by the computer system. The purpose is to adjust the preset abnormal data identification model based on the first prior distribution heat map and the live monitoring information to be processed to obtain a converged abnormal data identification model.

[0207] In step S510, the liveness monitoring information to be processed is inferred based on a preset anomaly data identification model to obtain a first inference distribution heatmap. The preset anomaly data identification model is a pre-built model designed to identify the distribution locations of anomaly data in the liveness monitoring information. This model may be based on various machine learning algorithms, such as neural networks.

[0208] Taking cardiovascular system monitoring as an example, suppose the in vivo monitoring information to be processed includes data such as heart rate, blood pressure, and electrocardiogram (ECG) signals. After receiving this data, the pre-defined abnormal data identification model first preprocesses the data. For heart rate data, normalization may be performed to map the heart rate value to a specific interval, such as [0, 1]. Suppose the original heart rate data is [60, 65, 70, 80, 90, 100, 110, 120, 110, 100, 90, 80, 75, 70, 65] (unit: beats / minute), after normalization, it may become [0.2, 0.25, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.7, 0.6, 0.5, 0.4, 0.35, 0.3, 0.25].

[0209] For blood pressure data, we assume systolic and diastolic pressures are processed separately. The original systolic pressure data is [100, 110, 120, 130, 140, 150, 160, 150, 140, 130, 120, 110, 105, 100, 95] (unit: mmHg), which, after similar normalization, becomes [0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.6, 0.5, 0.4, 0.3, 0.2, 0.15, 0.1, 0.05]; the original diastolic pressure data is [60, 65, 70, 75, 80, 85, 90, 85, 80, 75, 70, 65, 62, 60,

[58] , after normalization becomes [0.1, 0.15, 0.2, 0.25, 0.3, 0.35, 0.4, 0.35, 0.3, 0.25, 0.2, 0.15, 0.12, 0.1, 0.08]. For electrocardiogram signals, due to their complexity, the model may extract some key features, such as the width of the QRS complex and the ST segment shift. Assuming the extracted QRS group width feature value is [0.08, 0.09, 0.1, 0.11, 0.12, 0.11, 0.1, 0.09, 0.08, 0.09, 0.1, 0.11, 0.1, 0.09, 0.08] (unit: seconds), after normalization, it may become [0.2, 0.3, 0.4, 0.5, 0.6, 0.5, 0.4, 0.3, 0.2, 0.3, 0.4, 0.5, 0.4, 0.3, 0.2].

[0210] The model then inputs this preprocessed data into its internal neural network structure. The input layer of the neural network receives these normalized heart rate, blood pressure, and electrocardiogram (ECG) signal features. The hidden layers perform complex nonlinear transformations on these input data. For example, a simple hidden layer might contain multiple neurons, each of which performs a weighted summation with a bias term before outputting an activation function (such as ReLU). Assuming a hidden layer has 5 neurons, for the input value of heart rate, the calculation after the first neuron might be... ,in It is the weight of the input data to the first neuron. It is the normalized value of heart rate. It is a bias term.

[0211] After processing by the hidden layer, the output layer outputs results regarding the location of outlier data. This result is presented as a first inference distribution heatmap. For example, in a heatmap representing the probability of outlier data distribution at different time points within a 24-hour period, if the number of neurons in the output layer corresponds to 24 time points, then each neuron outputs a value representing the probability of outlier data occurring at that time point. Assuming the output of the first inference distribution heatmap has values ​​of [0.1, 0.12, 0.15, 0.2, 0.25, 0.3, 0.35, 0.4, 0.35, 0.3, 0.25, 0.2, 0.15, 0.12, 0.1, 0.08, 0.05, 0.03, 0.02, 0.01, 0.01, 0.02, 0.03, 0.05], a higher value for a given time point indicates a greater probability of abnormal data (such as abnormal heart rate, blood pressure, or abnormal electrocardiogram signal) occurring at that time point.

[0212] In step S520, after obtaining the first inference distribution heatmap and the first prior distribution heatmap, the computer system needs to calculate the error between the two to obtain the first error result of the preset anomaly data identification model. The first prior distribution heatmap is obtained in step S400 based on the distribution location inference markers, and it reflects the prior probability distribution of the anomaly data distribution location obtained based on the inference in the previous steps; while the first inference distribution heatmap is the result obtained by the preset anomaly data identification model based on its own algorithm to infer the data.

[0213] Continuing with the example of cardiovascular system monitoring, let's assume the values ​​of the first prior heatmap are [0.15, 0.18, 0.2, 0.22, 0.25, 0.32, 0.38, 0.42, 0.38, 0.32, 0.25, 0.22, 0.18, 0.15, 0.12, 0.1, 0.08, 0.06, 0.04, 0.03, 0.02, 0.02, 0.03, 0.04].

[0214] The mean squared error (MSE) can be used to calculate the error. The formula for calculating the mean squared error is: , where n is the number of elements in the heatmap (24 in this example, representing 24 time points). It is the i-th element in the first prior distribution heatmap. It is the i-th element in the first reasoning distribution heatmap.

[0215] For the first time point, For the second time point, Similarly, calculate the sum of squared errors for all 24 time points, and then divide by 24 to obtain the mean square error.

[0216] Besides mean squared error, other error measures can be used, such as mean absolute error (MAE), which is calculated using the following formula: The computer system calculates these error metrics to obtain the first error result of the preset anomaly data identification model. This result reflects the degree of difference between the model's current inference result and the prior distribution heatmap obtained based on the previous steps.

[0217] In step S530, after obtaining the first error result, the computer system adjusts the preset anomaly identification model based on this result to obtain a converged anomaly identification model. The adjustment process aims to reduce the model's error and make the model's output closer to the actual distribution of anomaly data.

[0218] Assuming the pre-defined anomaly detection model is a neural network model, the tuning process mainly involves adjusting the weights and biases of the neural network. Taking a simple neural network structure as an example, assume the input layer has 3 nodes (corresponding to normalized data of heart rate, blood pressure, and electrocardiogram signals), the hidden layer has 5 neurons, and the output layer has 24 nodes (corresponding to the probability distribution of anomaly data at 24 time points).

[0219] In the backpropagation algorithm, the gradient of each weight and bias term is calculated based on the first error result. For example, for the connection weights from the input layer to the hidden layer... Calculate its gradient (If mean squared error is used as the first error result). This gradient represents the weights. The direction and magnitude of the influence of minute changes on the error.

[0220] Suppose that during the calculation process, it is discovered that the weights of the heart rate input to the first hidden layer neuron are... Its gradient is positive, which means that increasing A value that leads to increased error will cause the calibration process to proceed according to a certain learning rate (e.g., ...). Decrease The value, the new Value .

[0221] Similarly, the weights within the hidden layers, the weights from the hidden layers to the output layer, and the biases of each neuron are adjusted based on their corresponding gradients. Through multiple iterations of this tuning process, the model's weights and biases are continuously adjusted based on the new initial error results until the model's error reaches an acceptable range or no longer decreases significantly.

[0222] For example, after multiple iterations, the model's first error result (such as mean squared error) decreased from the initial 0.05 to 0.01. At this point, the preset anomaly identification model converged after adjustment, resulting in a converged anomaly identification model. This converged model can more accurately identify the distribution location of abnormal data in the liveness monitoring information to be processed, thereby improving the accuracy and reliability of the entire liveness monitoring information processing method. Steps S510-S530, through the reasoning, error calculation, and adjustment process of the preset anomaly identification model, gradually optimize the model, ultimately obtaining the converged anomaly identification model.

[0223] As one implementation method, step S510, based on a preset abnormal data identification model, infers the location of abnormal data in the liveness monitoring information to be processed to obtain a first inference distribution heatmap, which may include:

[0224] Step S511: Based on the preset abnormal data identification model, extract the representation array of the live monitoring information to be processed to obtain the initial array representation of the live monitoring information to be processed;

[0225] Step S512: Based on the preset abnormal data identification model, perform a convolution operation on the initial array representation of the liveness monitoring information to be processed to obtain the convolution array representation of the liveness monitoring information to be processed;

[0226] Step S513: Perform residual connection operation based on the initial array representation and the convolution array representation to obtain the first inference distribution heat map of the liveness monitoring information to be processed.

[0227] In step S511, when the computer system performs inference based on the preset anomaly data identification model for the liveness monitoring information to be processed, it first performs representation array extraction. Representation array extraction is a feature extraction process. When processing liveness monitoring information, this information is often complex and contains multiple features. The computer system needs to extract an array representation that can represent its essential features from this complex data. For example, suppose the liveness monitoring information to be processed is monitoring data about cardiac activity. This data may include the changes of multiple parameters such as heart rate, electrocardiogram waveform, and cardiac systolic and diastolic pressure over time. The computer system will analyze these parameters, select key features, and combine them into an initial array representation. If the peak and trough values ​​of the heart rate, electrocardiogram waveform, and the values ​​of cardiac systolic and diastolic pressure at specific moments are taken as key features, then these feature values ​​will be organized into an initial array. This initial array can reflect the core features of the liveness monitoring information to be processed to a certain extent, providing a foundation for subsequent model inference operations. For the preset anomaly data identification model, this initial array representation is an important component of the model input. The model identifies the location of abnormal data distribution based on learning from large amounts of data, and this initial array representation presents the liveness detection information to be processed in a form suitable for the model to handle. It's like a translation process, "translating" the raw, complex liveness detection information into an array form that the model can understand, enabling subsequent operations. If the model is a neural network, each element of this initial array is equivalent to the input value of a neuron, which the neural network processes based on its internal weights and connections.

[0228] In step S512, after obtaining the initial array representation of the liveness monitoring information to be processed, the computer system performs a convolution operation. Convolution is a mathematical operation widely used in image processing, signal processing, and other fields, and it also plays a significant role in this liveness monitoring information processing. Taking the previously mentioned cardiac activity monitoring data as an example, assume that the elements in the initial array representation represent values ​​of different time points and different features to a certain extent. The convolution operation can be seen as a process of extracting local features from this data. For example, a specific convolution kernel (which can be a small array, such as a 3×3 array, where the elements are pre-defined weight values) is used to perform calculations by sliding across the initial array. This convolution kernel acts like a filter, extracting feature combinations from different local regions in the initial array. For instance, the convolution kernel might emphasize the correlation between certain specific heart rate and ECG waveform combinations, and by sliding the calculations across the initial array, these correlations are reflected in the convolution array representation in a new numerical form.

[0229] This convolution operation transforms the initial array representation into a convolutional array representation. During this process, the data features are further refined and transformed. Some subtle, localized features in the initial array representation may be amplified or suppressed after the convolution operation. For cardiac activity monitoring data, some subtle changes in heart rate and ECG waveforms that are not easily identifiable in the initial array may become more apparent in the convolutional array representation after the convolution operation. This helps the model better capture key information in the data because the convolutional array representation focuses more on specific patterns and relationships in the data, which may be closely related to the distribution location of anomalous data.

[0230] In step S513, after obtaining the initial array representation and the convolutional array representation, the computer system performs a residual connection operation on these two to obtain the first inference distribution heatmap of the liveness monitoring information to be processed. The basic principle of the residual connection operation is to add or combine the input information (here, the initial array representation) with the processed information (the convolutional array representation) in a specific way. Taking the previous example, suppose the initial array representation contains the basic features of the original cardiac activity monitoring data, while the convolutional array representation contains the local features extracted through the convolution operation. The residual connection operation is like fusing these two different levels of feature information. For example, for a specific time point and feature value, the heart rate value in the initial array representation and the local feature values ​​related to that heart rate value, emphasized by the convolution operation in the convolutional array representation, will be added or otherwise combined. The purpose of this operation is to allow the model to better retain the information in the original data while utilizing the processed feature information.

[0231] The first inference distribution heatmap obtained through the residual join operation is the final output of step S510. This heatmap reflects, to some extent, the preliminary inference result of the computer system based on the preset abnormal data identification model in the live monitoring information to be processed regarding the distribution location of abnormal data. In the example of cardiac activity monitoring, this heatmap may show higher heat values ​​in certain areas, indicating a higher probability of abnormal data distribution in these areas (corresponding to specific heart rate ranges, ECG waveform intervals, or cardiac pressure intervals, etc.). This inference result is based on the combined effect of the preceding representation array extraction, convolution operation, and residual join operation. Representation array extraction provides basic data features, convolution operation further mines local features, and residual join operation merges the two to obtain this heatmap that reflects the preliminary inference of the distribution location of abnormal data, providing an important basis for subsequent steps, such as calculating error results and adjusting the model.

[0232] As one implementation method, step S530, adjusting the preset abnormal data identification model based on the first error result to obtain a converged abnormal data identification model, may include:

[0233] Step S531: Based on the abnormal data distribution results of the reference liveness monitoring information, obtain the second prior distribution heat map of the abnormal data distribution location in the reference liveness monitoring information;

[0234] Step S532: Based on the preset abnormal data identification model, reason about the reference liveness monitoring information to obtain a second reasoning distribution heat map of the distribution location of abnormal data in the reference liveness monitoring information;

[0235] Step S533: Based on the second inference distribution heatmap and the second prior distribution heatmap, obtain the second error result of the preset abnormal data identification model;

[0236] Step S534: Based on the first error result and the second error result, adjust the preset abnormal data identification model to obtain a converged abnormal data identification model.

[0237] The purpose of step S531 in the computer system is to obtain a second prior heatmap of the abnormal data distribution locations in the reference liveness monitoring information based on the abnormal data distribution results of the reference liveness monitoring information. This second prior heatmap is a probabilistic description of the abnormal data distribution in the reference liveness monitoring information, reflecting prior information about the possible locations of abnormal data in the reference data. This prior information is crucial for tuning the preset abnormal data identification model because it provides the model with an expectation of the abnormal data distribution under normal conditions (based on the reference data), thereby helping the model adjust its parameters to better adapt to new data (the liveness monitoring information to be processed). Assume the reference liveness monitoring information is monitoring data about human blood glucose levels, which is collected over a period of time through an implantable medical device. The computer system first obtains the abnormal data distribution results in the reference liveness monitoring information, for example, the distribution of abnormally high or low blood glucose levels during certain time periods (such as within two hours after a meal, early morning, etc.) or under specific physical states (such as after exercise, during emotional fluctuations, etc.). Then, the computer system constructs a second prior heatmap based on these abnormal data distribution results. For example, if abnormal blood sugar spikes are more frequent within two hours after a meal, the corresponding area on the heatmap will show a higher heat value (indicating a higher probability); conversely, if abnormal blood sugar spikes are less frequent during sleep at night, the corresponding area will show a lower heat value. This heatmap can be in the form of a two-dimensional array, with one dimension representing time and the other representing other relevant factors such as physical condition. Each element in the array represents the probability value of abnormal data distribution under the corresponding conditions.

[0238] In step S532, when the computer system infers the reference liveness monitoring information based on the preset anomaly data identification model, its purpose is to obtain a second inference distribution heatmap of the distribution location of anomaly data in the reference liveness monitoring information. The preset anomaly data identification model is a pre-constructed model that includes certain algorithms and parameter structures, used to analyze the input liveness monitoring information and infer the distribution location of anomaly data. When inferring the reference liveness monitoring information, the model processes the data according to its internal computational logic.

[0239] Continuing with the example of blood glucose monitoring data, let's assume the pre-set abnormal data identification model is a neural network-based model. The computer system inputs reference live surveillance information (such as the sequence of blood glucose values ​​changing over time and related physical condition indicators) into the model. The neurons in the model process this input data layer by layer according to pre-learned weights and activation functions. For example, the first layer of neurons may perform preliminary feature extraction on the input blood glucose values ​​and physical condition indicators, such as identifying the trend of blood glucose value changes and the preliminary correlation between physical condition and blood glucose fluctuations. As the data propagates through the neural network, subsequent neurons further analyze the complex relationships between these features, ultimately obtaining an inference result about the distribution location of abnormal data in the reference live surveillance information. This result is presented in the form of a second inference distribution heatmap. Similar to the second prior distribution heatmap in step S531, this heatmap is also in the form of a two-dimensional array, where the heat value represents the probability of the abnormal data distribution inferred by the model under different conditions.

[0240] The computer system executes step S533 to obtain a second error result for the preset anomaly identification model based on the second inference distribution heatmap and the second prior distribution heatmap. This error result reflects the degree of difference between the model's inference result (second inference distribution heatmap) and the prior expectation based on reference liveness monitoring information (second prior distribution heatmap). By calculating this error, the computer system can evaluate the model's performance on the reference liveness monitoring information and thus determine the direction and magnitude of model adjustments.

[0241] To address computational errors, computer systems can employ various methods, such as the mean squared error (MSE) method. Suppose that an element in the second prior heatmap (corresponding to the probability distribution of outlier data at a specific time and under a specific physical condition) is... The element at the corresponding position in the second inference distribution heatmap is... Then the mean square error Where n and m are the number of elements in the heatmap in two dimensions, respectively. Taking blood glucose monitoring data as an example, if during a certain time period (e.g., 10-11 am), the probability of abnormally high blood glucose in the second prior distribution heatmap is... The corresponding probability in the second inference distribution heatmap derived by the model is... These differences are reflected in the overall mean squared error calculation. This error result reflects the degree to which the model's inference about the location of abnormal data distribution based on reference liveness monitoring information deviates from the actual prior situation.

[0242] In step S534, the computer system adjusts the preset anomaly data identification model based on the first error result and the second error result to obtain a converged anomaly data identification model. The first error result is calculated based on the liveness monitoring information to be processed, and the second error result is calculated based on the reference liveness monitoring information. These two error results reflect the model's performance from different perspectives. The purpose of adjusting the model is to enable it to better adapt to the data and reduce errors in identifying the distribution location of anomalies. Assume that the preset anomaly data identification model is a model based on the gradient descent algorithm. The computer system will adjust the model's parameters based on the first error result and the second error result. For example, for a certain weight parameter w in the model, if the first error result indicates that the model has a large error in judging the distribution location of anomalies for a certain feature (such as a certain ECG waveform feature in cardiac monitoring) on ​​the liveness monitoring information to be processed, and the second error result also shows that there is a deviation in the processing of similar features in the reference liveness monitoring information, then the computer system will adjust the weight w according to the direction (whether the predicted probability is too high or too low) and magnitude of the error. If the error indicates that the predicted probability is too high, the computer system will decrease the value of w according to a certain learning rate α (e.g., (where E is the error function). By adjusting multiple parameters in the model in this way, the model gradually optimizes in the direction of reducing errors, and finally obtains a converged anomaly identification model. This converged model is more accurate in identifying and marking the distribution location of anomaly data in the liveness monitoring information to be processed, because it has comprehensively considered the error information in the data to be processed and the reference data and made corresponding adjustments.

[0243] It should be noted that the in vivo monitoring information processing method based on implantable medical control devices provided in this application only marks the data as training data for the subsequent convergence abnormal data identification model, and does not involve the generation of specific disease diagnosis and treatment suggestions.

[0244] 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.

[0245] This invention provides a computer system comprising: one or more processors; a memory; and 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 programs are executed by the processors, they implement the method provided in this application.

[0246] The above description is only a partial embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A living body monitoring information processing method based on an implanted medical control device, characterized by, The method comprises: obtaining a reference living body monitoring information library and a to-be-processed living body monitoring information library, wherein the reference living body monitoring information library comprises one or more reference living body monitoring information and abnormal data distribution results of the reference living body monitoring information, and the to-be-processed living body monitoring information library comprises a plurality of to-be-processed living body monitoring information; the living body monitoring information in the reference living body monitoring information library and the to-be-processed living body monitoring information library is collected based on an implantable medical regulation and control device; performing distance calculation on abnormal data array representation of the reference living body monitoring information and monitoring information array representation of the to-be-processed living body monitoring information according to the abnormal data distribution results, to obtain a similarity score of the to-be-processed living body monitoring information; performing reasoning according to the similarity score to obtain a distribution position reasoning mark of abnormal data in the to-be-processed living body monitoring information; specifically, performing abnormal data distribution information reasoning according to the similarity score to obtain a first abnormal data distribution information probability density of the to-be-processed living body monitoring information; performing step-by-step scanning on the first abnormal data distribution information probability density to obtain a regional peak response intensity of each first scanning region of the first abnormal data distribution information probability density; if the regional peak response intensity of the first scanning region is greater than a first set threshold, a first distribution information corresponding to the regional peak response intensity of the first scanning region is determined as the distribution position reasoning mark; if the regional peak response intensity of the first scanning region is not greater than the first set threshold, performing reasoning on the to-be-processed living body monitoring information based on a convergent abnormal data recognition model to obtain a mark result of the distribution position of abnormal data in the to-be-processed living body monitoring information; obtaining a first prior distribution heat map of the distribution position of abnormal data in the to-be-processed living body monitoring information according to the distribution position reasoning mark; performing adjustment on a preset abnormal data recognition model according to the first prior distribution heat map and the to-be-processed living body monitoring information to obtain a convergent abnormal data recognition model, wherein the convergent abnormal data recognition model is used for performing distribution position recognition marking of abnormal data on to-be-processed living body monitoring information.

2. The living body monitoring information processing method based on the implantable medical regulation apparatus according to claim 1, wherein The reasoning on the to-be-processed living body monitoring information based on the convergent abnormal data recognition model to obtain the mark result of the distribution position of abnormal data in the to-be-processed living body monitoring information comprises: performing reasoning on to-be-processed living body monitoring information based on the convergent abnormal data recognition model to obtain a second abnormal data distribution information probability density of the to-be-processed living body monitoring information; performing step-by-step scanning on the second abnormal data distribution information probability density to determine a regional peak response intensity of each second scanning region of the second abnormal data distribution information probability density; if the regional peak response intensity of the second scanning region is greater than a second set threshold, a second distribution information corresponding to the regional peak response intensity of the second scanning region is determined as the mark result of the distribution position of abnormal data in the to-be-processed living body monitoring information.

3. The living body monitoring information processing method based on an implantable medical regulation apparatus according to Claim 1, wherein The convergent abnormal data identification model inferences the to-be-processed living body monitoring information to obtain a marking result of an abnormal data distribution position in the to-be-processed living body monitoring information, including: Inference is performed on the to-be-processed living body monitoring information based on the convergent abnormal data identification model to obtain a second abnormal data distribution information probability density of the to-be-processed living body monitoring information; The distribution information with the maximum abnormal data distribution information probability is obtained according to the second abnormal data distribution information probability density, and is determined as the marking result of the abnormal data distribution position in the to-be-processed living body monitoring information.

4. The living body monitoring information processing method based on the implantable medical regulation apparatus according to Claim 1, wherein The reference living body monitoring information library includes first reference living body monitoring information and second reference living body monitoring information, and distance calculation is performed on abnormal data array representations of the reference living body monitoring information and monitoring information array representations of the to-be-processed living body monitoring information according to the abnormal data distribution result of the to-be-processed living body monitoring information to obtain a similarity score of the to-be-processed living body monitoring information, including: The abnormal data array representation of the first reference living body monitoring information is obtained according to the abnormal data distribution result of the first reference living body monitoring information, and the abnormal data array representation of the second reference living body monitoring information is obtained according to the abnormal data distribution result of the second reference living body monitoring information; The monitoring information array representation of the to-be-processed living body monitoring information is obtained by performing array extraction on the to-be-processed living body monitoring information; Distance calculation is performed on the abnormal data array representation of the first reference living body monitoring information and the monitoring information array representation to obtain a first similarity measurement value between the first reference living body monitoring information and the to-be-processed living body monitoring information; Distance calculation is performed on the abnormal data array representation of the second reference living body monitoring information and the monitoring information array representation to obtain a second similarity measurement value between the second reference living body monitoring information and the to-be-processed living body monitoring information; The first similarity measurement value and the second similarity measurement value are integrated to obtain the similarity score; The convergent abnormal data identification model is obtained by calibrating a preset abnormal data identification model according to the first prior distribution heat map and the to-be-processed living body monitoring information, including: Inference is performed on the to-be-processed living body monitoring information based on the preset abnormal data identification model to obtain a first inference distribution heat map of an abnormal data distribution position in the to-be-processed living body monitoring information; A first error result of the preset abnormal data identification model is obtained according to the first inference distribution heat map and the first prior distribution heat map; The convergent abnormal data identification model is obtained by calibrating the preset abnormal data identification model according to the first error result.

5. The living body monitoring information processing method based on the implantable medical regulation apparatus according to claim 4, wherein The convergent abnormal data identification model is obtained by calibrating a preset abnormal data identification model according to the first prior distribution heat map and the to-be-processed living body monitoring information, including: An initial array representation of the to-be-processed living body monitoring information is obtained by performing array extraction on the to-be-processed living body monitoring information based on the preset abnormal data identification model; performing convolution operation on the initial array representation of the to-be-processed living body monitoring information based on the preset abnormal data identification model to obtain a convolution array representation of the to-be-processed living body monitoring information; performing residual connection operation on the initial array representation and the convolution array representation to obtain a first inference distribution heat map of the to-be-processed living body monitoring information.

6. The living body monitoring information processing method based on the implantable medical regulation apparatus according to claim 4, wherein The adjusting and calibrating the preset abnormal data identification model based on the first error result to obtain a converged abnormal data identification model comprises: obtaining a second prior distribution heat map of abnormal data distribution positions in the reference living body monitoring information based on the abnormal data distribution result of the reference living body monitoring information; performing inference on the reference living body monitoring information based on the preset abnormal data identification model to obtain a second inference distribution heat map of the abnormal data distribution positions in the reference living body monitoring information; obtaining a second error result of the preset abnormal data identification model based on the second inference distribution heat map and the second prior distribution heat map; adjusting and calibrating the preset abnormal data identification model based on the first error result and the second error result to obtain a converged abnormal data identification model.

7. The living body monitoring information processing method based on an implantable medical regulation apparatus according to any one of claims 1 to 6, characterized by, The obtaining a first prior distribution heat map of abnormal data distribution positions in the to-be-processed living body monitoring information based on the distribution position inference label comprises: performing binary coding on the to-be-processed living body monitoring information based on the distribution position inference label to obtain coded monitoring information of the to-be-processed living body monitoring information; performing smoothing filtering processing on the coded monitoring information to obtain the first prior distribution heat map of the abnormal data distribution positions in the to-be-processed living body monitoring information.

8. A computer system, characterized by comprise: 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 one or more processors, the method according to any one of claims 1-7 is implemented.

Images