Monitoring methods and equipment

Abstract

An embodiment of the present invention provides a monitoring method and device. In one embodiment, the method comprises: acquiring physiological signals; detecting the acquired physiological signals using multiple artificial intelligence alarm sub-models, each artificial intelligence alarm sub-model using the physiological signals as input to detect a specific type of abnormality and outputting an alarm message when an abnormality is detected; synthesizing the alarm messages output by the multiple artificial intelligence alarm sub-models; and outputting the synthesized alarm message. The method of this embodiment of the present invention utilizes different artificial intelligence alarm sub-models for detecting different types of abnormalities.

Description

Technical Field

[0001] The present invention relates to the technical field of medical equipment, and in particular to a monitoring method and equipment. Background Art

[0002] Monitoring devices collect and analyze patients' physiological signals, alerting medical staff when abnormalities occur. Traditional monitoring devices are typically based on threshold-based expert systems. These thresholds are often statistically determined based on clinical experience, lacking individuality and comprehensive considerations. With the rapid development of machine learning, traditional threshold-based expert systems are gradually being replaced by solutions based on artificial intelligence.

[0003] Current AI-based solutions use supervised training of AI models using a large number of physiological signal samples labeled with alarm categories. The trained AI models then detect the patient's physiological signals, identify abnormalities, and trigger alarms. However, different alarm categories require different information, which can interfere with each other during model training, resulting in reduced alarm accuracy. Therefore, the alarm accuracy of existing AI-based solutions needs to be improved. Summary of the Invention

[0004] The present invention mainly provides a monitoring method and device, which are used to solve the problem of low alarm accuracy of existing monitoring devices.

[0005] According to the first aspect, an embodiment provides a monitoring method, comprising:

[0006] Acquire physiological signals;

[0007] The acquired physiological signals are detected by multiple artificial intelligence alarm sub-models. Each artificial intelligence alarm sub-model uses the physiological signal as input to detect a type of abnormal situation and outputs an alarm message when an abnormal situation is detected.

[0008] Integrate the alarm information output by multiple artificial intelligence alarm sub-models;

[0009] Output the integrated alarm information.

[0010] According to the second aspect, an embodiment provides a monitoring device, including:

[0011] A signal acquisition circuit, which acquires physiological signals using sensor accessories connected to the patient;

[0012] Output module, used to output alarm information;

[0013] Memory, used to store programs;

[0014] A processor is used to implement the monitoring method as described in any one of the first aspects by executing the program stored in the memory.

[0015] According to a third aspect, an embodiment provides a computer-readable storage medium, comprising a program, wherein the program can be executed by a processor to implement the method described in any embodiment herein.

[0016] The monitoring method and device provided by the embodiments of the present invention acquire physiological signals; detect the acquired physiological signals using multiple artificial intelligence alarm sub-models. Each artificial intelligence alarm sub-model uses the physiological signals as input to detect a specific type of abnormality and outputs an alarm message when an abnormality is detected; the alarm information output by the multiple artificial intelligence alarm sub-models is integrated; and the integrated alarm message is output. This implements the use of different artificial intelligence alarm sub-models for detection of different types of abnormalities, helping to improve the accuracy of the alarm information. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 A schematic diagram of the structure of a monitoring device provided in one embodiment;

[0018] Figure 2 A schematic structural diagram of a monitoring device provided in another embodiment;

[0019] Figure 3 A flowchart of a monitoring method provided in one embodiment;

[0020] Figure 4 A schematic diagram of determining an artificial intelligence sub-model based on the length of a physiological signal required to detect an abnormality, provided by one embodiment;

[0021] Figure 5 for Figure 4 Schematic diagram of the structure of the artificial intelligence alarm sub-model;

[0022] Figure 6 A schematic diagram of an alarm configuration interface provided by an embodiment;

[0023] Figure 7 A schematic diagram of a combined structure of multiple artificial intelligence alarm sub-models provided in one embodiment;

[0024] Figure 8 A schematic diagram of a combined structure of multiple artificial intelligence alarm sub-models provided in another embodiment;

[0025] Figure 9 This is a structural diagram of a monitoring device provided in yet another embodiment. DETAILED DESCRIPTION

[0026] The present invention will be further described in detail below by means of specific embodiments in conjunction with the accompanying drawings. Similar elements in different embodiments are numbered with associated similar elements. In the following embodiments, many detailed descriptions are provided to enable the present application to be better understood. However, those skilled in the art will readily appreciate that some of the features may be omitted in different circumstances, or may be replaced by other elements, materials, or methods. In some cases, some operations related to the present application are not shown or described in the specification. This is to avoid the core portion of the present application being overwhelmed by excessive descriptions, and for those skilled in the art, it is not necessary to describe these related operations in detail. They will fully understand the related operations based on the description in the specification and the general technical knowledge in the art.

[0027] In addition, the features, operations, or characteristics described in the specification may be combined in any appropriate manner to form various embodiments. Furthermore, the steps or actions in the method description may be reordered or adjusted in a manner readily apparent to those skilled in the art. Therefore, the various sequences in the specification and drawings are provided solely for the purpose of clearly describing a particular embodiment and are not intended to be mandatory, unless otherwise specified.

[0028] The serial numbers assigned to components herein, such as "first," "second," etc., are used solely to distinguish the objects being described and do not convey any sequential or technical meaning. References to "connection" and "coupling" herein, unless otherwise specified, include both direct and indirect connections (couplings).

[0029] like Figure 1 FIG. 1 shows a schematic diagram of the structure of a monitoring device 100 that can be used for multi-parameter monitoring. The monitoring device 100 may have an independent housing, and the housing panel may include a sensor interface area. The sensor interface area may integrate multiple sensor interfaces for connecting to external physiological parameter sensor accessories 111. The housing panel may also include a small LCD display area, a display 119, an input interface circuit 122, and an alarm circuit 120 (such as an LED alarm area). The monitoring device 100 may have an external communication and power interface 116 for communicating with and drawing power from a host computer. The monitoring device 100 may also support an external parameter module. By inserting the parameter module, a plug-in monitoring device 100 host computer can be formed as a part of the monitoring device 100, or it can be connected to the host computer via a cable, with the external parameter module being an external accessory of the monitoring device 100.

[0030] The internal circuit of the monitoring device 100 is placed in the housing. Figure 1As shown, the system includes a signal acquisition circuit 112 corresponding to at least two physiological parameters, a front-end signal processing circuit 113, and a main processor 115. The signal acquisition circuit 112 can be selected from an electrocardiogram (ECG) circuit, a respiratory circuit, a body temperature circuit, a blood oxygenation circuit, a non-invasive blood pressure (NIBP) circuit, an invasive blood pressure (IBP) circuit, and the like. Each of these signal acquisition circuits 112 is electrically connected to a corresponding sensor interface for electrically connecting to a sensor accessory 111 corresponding to a different physiological parameter. The output end of each signal acquisition circuit 112 is coupled to the front-end signal processing circuit 113. The communication port of the front-end signal processing circuit 113 is coupled to the main processor 115, and the main processor 115 is electrically connected to an external communication and power supply interface 116. The sensor accessories 111 and signal acquisition circuit 112 corresponding to the various physiological parameters can utilize conventional circuits. The front-end signal processing circuit 113 performs sampling and analog-to-digital conversion of the output signal of the signal acquisition circuit 112, and outputs control signals to control the measurement process of the physiological signals. These parameters include, but are not limited to, ECG, respiratory, body temperature, blood oxygenation, NIBP, and IBP parameters. The front-end signal processing circuit 113 can be implemented using a single-chip microcomputer or other semiconductor device, such as the PHLIPS LPC2136 or ADI's ADuC7021 mixed-signal single-chip microcomputer. It can also be implemented using an ASIC or FPGA. The front-end signal processing circuit 113 can be powered by an isolated power supply. The sampled data undergoes simple processing and packaging before being sent to the main processor 115 via an isolated communication interface. For example, the front-end signal processing circuit 113 can be coupled to the main processor 115 via an isolated power supply and communication interface 114. The front-end signal processing circuit 113 is powered by an isolated power supply because the DC / DC power supply, isolated by a transformer, serves to isolate the patient from the power supply equipment. The main objectives are: 1. Isolating the patient by floating the applied circuitry through the isolation transformer, minimizing patient leakage current; and 2. Preventing voltage or energy during defibrillation or electrosurgical operation from affecting intermediate circuit boards and components, such as the main control board (guaranteed by creepage distance and electrical clearance). Of course, the front-end signal processing circuit 113 can also be connected to the main processor 115 via a cable 124. The main processor 115 is used to complete the calculation of physiological parameters and send the parameter calculation results and waveforms to a host (such as a host with a display, a PC, a central station, etc.) through the external communication and power interface 116; the main processor 115 can be connected to the external communication and power interface 116 via a cable 125 for communication and / or power supply; the monitoring device 100 may also include a power supply and battery management circuit 117, which draws power from the host through the external communication and power interface 116 and supplies it to the main processor 115 after processing, such as rectification and filtering; the power supply and battery management circuit 117 can also monitor, manage, and provide power protection for the power obtained from the host through the external communication and power interface 116.The external communication and power interface 116 can be one or a combination of local area network interfaces consisting of Ethernet, Token Ring, Token Bus, and the fiber-distributed data interface (FDDI) that serves as the backbone of these three networks. It can also be one or a combination of wireless interfaces such as infrared, Bluetooth, Wi-Fi, and WMTS communication, or it can be one or a combination of wired data connection interfaces such as RS232 and USB. The external communication and power interface 116 can also be one or a combination of wireless data transmission interfaces and wired data transmission interfaces. The host can be any computer device such as the host of the monitoring device 100, an electrocardiograph, an ultrasound diagnostic device, or a computer. By installing the corresponding software, the monitoring device 100 can be formed. The host can also be a communication device, such as a mobile phone. The monitoring device 100 can transmit data to a mobile phone that supports Bluetooth communication via a Bluetooth interface, enabling remote data transmission. The main processor 115 is also used to detect the physiological signals collected by the signal acquisition circuit 112 and output an alarm message when an abnormality is detected. The alarm circuit 120 and display 119 can be used as output modules for outputting alarm information. For example, the generated alarm information can be displayed on the display 119, or the alarm circuit 120 can emit an audible alarm to provide a prompt. The memory 118 can store intermediate and final data of the monitoring device 100, as well as program instructions or code for execution by the main processor 115, etc. If the monitoring device 100 has a blood pressure measurement function, it can also include a pump valve drive circuit 121, which is used to perform inflation and deflation operations under the control of the main processor 115.

[0031] Figure 1 The monitoring device 100 shown is a multi-parameter monitoring device. The monitoring device 100 can also be a single physiological parameter monitoring device. Figure 2 The following is an example. The same content can be found in the above Figure 1 content.

[0032] like Figure 3 As shown, the embodiment of the present invention provides a monitoring method that can be applied to Figure 1 or Figure 2 In the monitoring equipment shown in the figure, the alarm accuracy of the monitoring equipment can be improved. Figure 3 As shown, the monitoring method provided in this embodiment may include:

[0033] S301. Acquire physiological signals.

[0034] The physiological signal in this embodiment can be the original signal collected by the signal acquisition circuit through the sensor accessory, or it can be a signal generated by performing universal preprocessing on the collected original signal. Universal preprocessing can include, for example, lead-off processing, saturation processing, filtering processing, and signal normalization processing. Among them, signal normalization processing can unify the sampling rate and resolution of the signal to preset values. At the same time, for multi-channel signals, the signals of each channel can be arranged according to the common clinical arrangement order. Taking the electrocardiogram signal as an example, its resolution can be uniformly adjusted to 200Lsb / mV, the sampling rate can be uniformly adjusted to 250Hz, and the leads can be arranged in the order of I\II\III\aVR\aVL\aVF\V1~V6. The physiological signal in this embodiment is a continuous physiological signal, rather than a discrete parameter value, which avoids the loss of information caused by the extraction process from the continuous physiological signal to the discrete parameter value, resulting in reduced alarm accuracy.

[0035] Physiological signals in this embodiment include but are not limited to electrocardiogram (ECG), respiration, body temperature, blood oxygen, and blood pressure. ECG signal acquisition systems include but are not limited to 3-lead, 5-lead, and 12-lead systems, and blood pressure acquisition systems include but are not limited to cuff-type blood pressure acquisition systems.

[0036] S302. Detect the acquired physiological signals respectively through multiple artificial intelligence alarm sub-models. Each artificial intelligence alarm sub-model uses the physiological signal as input to detect a type of abnormal situation and outputs alarm information when an abnormal situation is detected.

[0037] After acquiring physiological signals, multiple artificial intelligence alarm sub-models can be used to detect the acquired physiological signals. In this embodiment, each artificial intelligence alarm sub-model can be pre-trained based on physiological signals labeled with alarm names and confidence levels. Each artificial intelligence alarm sub-model is used to detect a type of abnormality, taking physiological signals as input and outputting an alarm message when an abnormality is detected. The alarm message may include the alarm name and corresponding confidence level.

[0038] Using a single model would be incapable of specifically identifying multiple types of abnormalities because different types of abnormalities require different information. For example, when detecting ventricular fibrillation, the focus is on the presence of the QRS complex; when detecting atrial fibrillation, the focus is on the presence of P and F waves. By using multiple AI alarm sub-models to detect various types of abnormalities separately, we can focus on different information in a targeted manner.

[0039] If a single model is used, the performance of identifying anomalies in rare samples will be poor. Since rare samples are relatively rare compared to other samples, it is difficult to guarantee the accuracy of a trained single model in identifying anomalies in rare samples. For example, a rare sample only accounts for 1% of the natural distribution in the database. Although the overall accuracy of the trained single model is over 98%, it may have incorrectly predicted all anomalies in this rare sample. In this embodiment, however, an artificial intelligence alarm sub-model can be trained separately for rare samples to detect anomalies in rare samples, thereby improving the accuracy of identifying anomalies in rare samples.

[0040] In this embodiment, the abnormal situation to be detected by each artificial intelligence alarm sub-model can be determined based on the length of the physiological signal required to detect the abnormal situation, and / or the preprocessing operation required to detect the abnormal situation, and / or the similarity of the waveform corresponding to the abnormal situation.

[0041] The length of the physiological signals required to detect abnormal conditions caused by different types of diseases is not consistent. For example, accurately identifying abnormal rhythms such as atrial fibrillation requires a long period of time, generally more than 30 seconds, while detecting ventricular premature beats only requires a few heartbeats. If both atrial fibrillation and ventricular premature beats are detected in one model, in order to ensure the accuracy of atrial fibrillation detection, it is necessary to input a physiological signal segment with a length of more than 30 seconds. However, an excessively long physiological signal segment contains too much redundant information for ventricular premature beat detection, which will increase the difficulty of ventricular premature beat detection and reduce the accuracy. Therefore, in an optional embodiment, the abnormal conditions to be detected by each artificial intelligence alarm sub-model can be determined based on the length of the physiological signal required to detect the abnormal condition. That is, for abnormal conditions with different required physiological signal lengths, different artificial intelligence sub-models can be used for detection.

[0042] Please refer to Figure 4 For alarm class 1 such as bigeminy, tripegem and polymorphic ventricular premature contractions (PVC), the optimal detection data length required is 5 seconds, and the artificial intelligence alarm sub-model 1 can be used to detect the abnormal situation of alarm class 1; for alarm class 2 such as atrial fibrillation, atrial flutter and irregular rhythm, the optimal detection data length required is 30 seconds, and the artificial intelligence alarm sub-model 2 can be used to detect the abnormal situation of alarm class 2; for alarm class n such as ventricular tachycardia, ventricular bradycardia and ventricular fibrillation, the optimal detection data length is 10 seconds, and the artificial intelligence alarm sub-model n can be used to detect the abnormal situation of alarm class n.

[0043] Each AI alarm sub-model can use the same structure or different structures. Figure 5 , showing Figure 4The structural diagram of artificial intelligence alarm sub-model 1 and artificial intelligence alarm sub-model 2. Figure 5 As shown in Figure 2, the length of the physiological signal input to the artificial intelligence alarm sub-model 2 is 5 times the length of the physiological signal input to the artificial intelligence alarm sub-model 1. Although both use convolutional neural networks, they use different structures for different alarm types.

[0044] Furthermore, the specific preprocessing operations required to identify different types of abnormal conditions are different. For example, for identifying atrial fibrillation, the morphology of the QRS wave is not critical; the focus is on the characteristics of the TQ wave. Therefore, in order to improve the accuracy of detecting atrial fibrillation, it is necessary to highlight the characteristics of the TQ wave and weaken the characteristics of the QRS wave. For identifying ventricular premature beats, the morphology of the QRS wave is crucial. Therefore, in order to improve the accuracy of detecting ventricular premature beats, it is necessary to highlight the morphological characteristics of the QRS wave and eliminate the variability of the TQ wave. Obviously, if a model is used to detect atrial fibrillation and ventricular premature beats at the same time, specific preprocessing operations cannot be performed. Therefore, in another optional embodiment, the abnormal conditions to be detected by each artificial intelligence alarm sub-model can be determined based on the preprocessing operations required to detect the abnormal conditions. That is, for abnormal conditions that require different preprocessing operations, different artificial intelligence sub-models can be used for detection, so that specific preprocessing can be performed in a targeted manner to highlight the key information required for identifying such abnormalities.

[0045] In another optional embodiment, the abnormal conditions to be detected by each AI alarm sub-model can be determined based on the similarity of the waveforms corresponding to the abnormal conditions. The waveforms corresponding to different abnormal conditions may have a certain degree of similarity. When this similarity exceeds a preset similarity threshold, they can be classified as the same abnormal condition. In other words, for multiple abnormal conditions with waveform similarity exceeding the preset similarity threshold, the same AI sub-model can be used for detection, thereby improving both detection accuracy and efficiency.

[0046] It should be noted that the abnormal conditions to be detected by each artificial intelligence alarm sub-model can also be determined based on two or all of the length of the physiological signal required to detect the abnormal condition, the preprocessing operation required to detect the abnormal condition, and the similarity of the waveform corresponding to the abnormal condition.

[0047] Training different AI alarm sub-models for different types of abnormal situations and detecting abnormal situations in a targeted manner will help improve the accuracy of alarms.

[0048] S303: Integrate the alarm information output by multiple artificial intelligence alarm sub-models.

[0049] Multiple artificial intelligence alarm sub-models may output multiple alarm messages, each of which may include an alarm name and a corresponding confidence level. In an optional embodiment, the alarm messages output by multiple artificial intelligence alarm sub-models can be combined based on the confidence level of the alarm messages output by each artificial intelligence alarm sub-model. For example, a confidence threshold can be preset or set according to user instructions, and only alarm messages with a confidence level higher than the confidence threshold are output. It should be noted that the same confidence threshold can be set for all alarm messages, or different confidence thresholds can be set for different alarm messages.

[0050] Furthermore, the priority of the alarm information output by each artificial intelligence alarm sub-model can also be determined. Specifically, the priority of each artificial intelligence alarm sub-model can be preset or set according to user instructions, and the priority of the alarm information is determined according to the priority of the artificial intelligence alarm sub-model that outputs the alarm information. The priority may include, for example, high, medium and low. In another optional embodiment, the alarm information output by multiple artificial intelligence alarm sub-models can also be integrated according to the priority of the alarm information output by each artificial intelligence alarm sub-model. For example, in certain specific scenarios, such as first aid situations, only high-priority alarm information is output; or, during the duration of the high-priority alarm information, the occurrence of low-priority alarms will not be displayed.

[0051] Optionally, the alarm information can be integrated according to a preset or user-set refractory period. That is, after a certain alarm information is output, a timer starts, and if the timer does not exceed the refractory period, no similar alarm is issued.

[0052] For alarm messages output by multiple AI alarm sub-models, a pre-trained value index model can be used to determine the value index of each alarm message and sort the alarm messages based on the value index. The value index model can take as input physiological signal waveform data and its signal quality index, alarm name and confidence level, alarm priority, and alarm messages from several time periods before and after the alarm message. The output is the value index of the current alarm message.

[0053] S304: Output the integrated alarm information.

[0054] After synthesizing the alarm information output by multiple AI alarm sub-models, the synthesized alarm information can be output through the monitoring device's output modules, such as displays, speakers, and signal lights. For example, the alarm name and corresponding confidence level can be displayed on the display; the alarm name and corresponding confidence level can be played through the speaker; and different alarm information can be prompted by different signal lights.

[0055] The monitoring method provided in this embodiment uses multiple artificial intelligence alarm sub-models to detect acquired physiological signals. Each artificial intelligence alarm sub-model uses physiological signals as input and is used to detect a specific type of abnormality. When an abnormality is detected, it outputs an alarm message, which includes the alarm name and corresponding confidence level. The alarm messages output by the multiple artificial intelligence alarm sub-models are then integrated and output. Using different artificial intelligence alarm sub-models for detection of different types of abnormalities helps improve the accuracy of the alarm information.

[0056] On the basis of the above embodiment, the method provided in this embodiment further includes: generating an alarm configuration interface so that the user can configure the alarm according to scenario requirements and personal operating habits.

[0057] In an optional embodiment, the alarm configuration interface can display the working status of each artificial intelligence alarm sub-model, and the working status includes an enabled state and a disabled state; and the working status of the corresponding artificial intelligence alarm sub-model can be updated in response to the user's operation on the working status. For example, an enable state button can be set for each artificial intelligence alarm sub-model in the alarm configuration interface, and the user can click the button corresponding to each sub-model to selectively turn on or off the corresponding sub-model according to needs. For example, in an emergency situation, only the sub-model for detecting rhythm analysis abnormalities can be turned on, while the sub-model for detecting morphological analysis abnormalities can be turned off, because at this time, due to the influence of motion interference and the like, the accuracy of the refined morphological analysis is greatly reduced and the number of false alarms increases.

[0058] In another optional embodiment, the alarm configuration interface may also display the alarm confidence thresholds for enabled AI alarm sub-models; in response to user manipulation of the alarm confidence thresholds, the alarm confidence thresholds for the corresponding AI alarm sub-models are updated. For example, a lower confidence limit setting box may be set for each AI alarm sub-model in the alarm configuration interface, allowing the user to enter an alarm confidence threshold for the enabled AI alarm sub-models through the lower confidence limit setting box. Each AI alarm sub-model will not output an alarm result for anomalies where the detected alarm confidence is lower than the user-set lower confidence limit, effectively avoiding a large number of false alarms.

[0059] In another optional embodiment, the alarm configuration interface can also display a combination structure of multiple AI alarm sub-models; and the combination structure of the multiple AI alarm sub-models can be updated in response to user operations on the combination structure. Initially, the combination structure of the multiple AI alarm sub-models can adopt a preset structure, and the user can redefine the combination structure through the visual interface, thereby improving the convenience of user model management.

[0060] For specific alarm configuration interface, please refer to Figure 6 As shown. Among them, the confidence lower limit of sub-model 1 in the enabled state is 75%, that is, sub-model 1 only outputs alarm information with a confidence level higher than 75%. The combined structure of multiple artificial intelligence alarm sub-models can be a serial structure, a parallel structure or a hybrid structure. Figure 6 The hybrid structure is shown in the figure. The following describes in detail the combined structure of the model and how to integrate alarm information for different combined structures through specific examples.

[0061] Please refer to Figure 7 The combined structure of multiple artificial intelligence alarm sub-models is a parallel structure, which can run multiple artificial intelligence alarm sub-models at the same time, and use the alarm information with the highest confidence and greater than the alarm confidence threshold among the alarm information output by multiple artificial intelligence alarm sub-models as the integrated alarm information.

[0062] Please refer to Figure 8 The combined structure of multiple AI alarm sub-models is a serial structure. For example, the AI ​​alarm sub-models can be organized into a serial structure according to their priority order. They can be run simultaneously or sequentially according to their priority order. The alarm information output by the multiple AI alarm sub-models with a confidence level greater than the alarm confidence threshold and the highest priority is used as the integrated alarm information.

[0063] And for Figure 6 The hybrid structure shown can be decomposed into a parallel structure and a serial structure to determine the integrated alarm information.

[0064] In order to adapt to different application scenarios and improve the flexibility of alarms, the monitoring method provided in this embodiment may also include: obtaining the correspondence between the application scenario and the working status of multiple artificial intelligence alarm sub-models; determining the current application scenario based on user operations; and starting or disconnecting the corresponding artificial intelligence alarm sub-model based on the working status of multiple artificial intelligence alarm sub-models corresponding to the current application scenario.

[0065] In different application scenarios, users expect the monitoring equipment to output different types of alarm information. For example, when doing a resting electrocardiogram, the user expects all possible abnormal conditions to be output; in first aid and defibrillation situations, the user expects only fatal abnormal conditions such as ventricular fibrillation to be output. Redundant alarm information will waste the user's energy and other resources. Therefore, the correspondence between the working status of each artificial intelligence alarm sub-model and the application scenario can be set in advance or according to user instructions. For the resting electrocardiogram application scenario, the working status of all artificial intelligence alarm sub-models is enabled; for the first aid application scenario, only the working status of the artificial intelligence sub-model for detecting ventricular fibrillation is enabled, and the working status of other artificial intelligence sub-models is disabled. The method provided in this embodiment can perform flexible alarms for different application scenarios.

[0066] Based on any of the above embodiments, a signal quality index of the physiological signal can also be determined based on the time-frequency domain characteristics of the physiological signal or using a pre-trained signal quality assessment model based on the original physiological signal. Specifically, before the acquired physiological signal is detected by multiple artificial intelligence alarm sub-models, the physiological signal can be analyzed to obtain the signal quality index of the physiological signal; and the process of detecting the physiological signal by the multiple artificial intelligence alarm sub-models can be controlled based on the signal quality index.

[0067] In an optional embodiment, the signal quality index of the physiological signal can be determined based on the time-frequency domain characteristics of the physiological signal. For example, the signal quality index of the physiological signal can be determined based on at least one of the amplitude, slope, and power spectrum of the physiological signal. Specifically, the signal quality index of the physiological signal can be determined according to the following formula:

[0068] δ=1-(α+β+2*γ) / 4;

[0069] Among them, δ represents the signal quality index, α represents the proportion of the amplitude of the physiological signal exceeding the preset amplitude range, and the preset amplitude range of the signal can be determined based on clinical experience and medical guidelines, and the proportion α exceeding the range can be counted, which can reflect the intensity of low-frequency noise in the saturation segment; β represents the proportion of the slope of the physiological signal exceeding the preset slope range, and the preset slope range of the signal can be determined based on the reasonable range of signal differential or high-order differential indicated by clinical experience and medical guidelines, and the proportion β exceeding the range can be counted, which can reflect the intensity of high-frequency noise interference; γ represents the power proportion of the frequency of the physiological signal exceeding the preset frequency range, and the spectrum-power distribution diagram of the signal can be calculated, and the preset frequency range of the physiological signal can be determined based on clinical experience and medical guidelines, and the power proportion γ exceeding the preset frequency range can be counted, which can comprehensively reflect the intensity of high and low frequency noise.

[0070] In another optional embodiment, a pre-trained signal quality assessment model can be used to determine the signal quality index of the physiological signal. Specifically, the physiological signal can be input into the pre-trained signal quality assessment model to obtain the signal quality index of the physiological signal, where the signal quality assessment model is trained based on the physiological signal labeled with the signal quality index.

[0071] A large amount of data containing noise of different intensities can be collected, and a signal quality index can be labeled for each data to establish a signal quality assessment database. The data in the signal quality assessment database has been normalized, and the label can be either a continuous percentage or a discrete sequence. Specifically, the quality of a real-time single-lead signal segment can be assessed every 1 second, and a data segment longer than 1 second is a weighted average of the signal quality indexes of all 1-second segments contained therein. In the case of multiple leads, the final signal quality index is the average of the signal quality indices on all leads. Then, a signal quality assessment model is trained based on the established signal quality assessment database, and the signal quality assessment model can be a deep convolutional model. When using the trained model for signal quality assessment, it is only necessary to input the normalized physiological signal segment into the model to output the signal quality index.

[0072] The signal quality index can be measured using either a continuous or discrete indicator. It can be a sequence describing quality levels, such as "Good Signal Quality," "Poor Signal Quality, Limited Use," or "Extremely Poor Signal Quality, Unusable," or "Level 1 Signal," "Level 2 Signal," "Level 3 Signal," or "Level 4 Signal." The quality index for extremely poor signals can be set to 0, while the quality index for normally usable signals can be set to 100. The remaining signal quality indices can vary continuously between 0 and 100.

[0073] After determining the signal quality index of the physiological signal, the signal quality index of the physiological signal can be outputted through the output module of the monitoring device. For example, the signal quality index can be displayed on a screen to indicate user confirmation and prompt the user to improve the signal quality.

[0074] After determining the signal quality index of the physiological signal, the process of detecting the physiological signal through multiple artificial intelligence alarm sub-models can also be controlled according to the signal quality index.

[0075] In an optional implementation, the working status of multiple artificial intelligence alarm sub-models can be determined based on the signal quality index; the acquired physiological signals are detected separately by the artificial intelligence alarm sub-models in the enabled state. Some artificial intelligence alarm sub-models have high requirements for signal quality. When the signal quality does not meet the requirements, the accuracy of the alarm will be greatly reduced, resulting in a large number of false alarms and missed alarms. Therefore, when the signal quality index is low, the working status of these artificial intelligence alarm sub-models can be set to a non-enabled state, and they will not participate in the detection of abnormal situations, thereby avoiding false alarms and missed alarms.

[0076] In another optional embodiment, the confidence level of the alarm information can be determined based on the signal quality index. The confidence level of the alarm information output by each artificial intelligence alarm sub-model is positively correlated with the signal quality index. In other words, the higher the signal quality, the higher the confidence level of the output alarm information.

[0077] The present invention also provides a monitoring device, see Figure 9 As shown. Figure 9 As shown, the monitoring device 90 provided in this embodiment may include: a signal acquisition circuit 901, an output module 902, a memory 903, a processor 904 and a bus 905. The bus 905 is used to realize the connection between various components.

[0078] The signal acquisition circuit 901 acquires physiological signals using sensor accessories connected to the patient;

[0079] Output module 902, used to output alarm information;

[0080] The memory 903 stores a computer program, which can implement the technical solution of any of the above method embodiments when executed by the processor 904.

[0081] This document is described with reference to various exemplary embodiments. However, those skilled in the art will recognize that changes and modifications may be made to the exemplary embodiments without departing from the scope of this document. For example, the various operational steps and components used to perform the operational steps may be implemented in different ways (e.g., one or more steps may be deleted, modified, or incorporated into other steps) depending on the specific application or considering any number of cost functions associated with the operation of the system.

[0082] Additionally, as will be appreciated by those skilled in the art, the principles of this disclosure may be embodied in a computer program product on a computer-readable storage medium pre-loaded with computer-readable program code. Any tangible, non-transitory computer-readable storage medium may be used, including magnetic storage devices (hard disks, floppy disks, etc.), optical storage devices (CD-ROMs, DVDs, Blu-ray discs, etc.), flash memory, and / or the like. These computer program instructions may be loaded onto a general-purpose computer, a special-purpose computer, or other programmable data processing device to form a machine, such that the instructions executed on the computer or other programmable data processing device can generate a device that implements a specified function. These computer program instructions may also be stored in a computer-readable memory, which can instruct the computer or other programmable data processing device to operate in a specific manner, such that the instructions stored in the computer-readable memory can form an article of manufacture that includes an implementation device that implements the specified function. The computer program instructions may also be loaded onto a computer or other programmable data processing device, causing the computer or other programmable device to execute a series of operational steps to produce a computer-implemented process, such that the instructions executed on the computer or other programmable device provide the steps for implementing the specified function.

[0083] The above examples are used to illustrate the present invention, which are only used to help understand the present invention and are not intended to limit the present invention. Those skilled in the art can make several simple deductions, modifications or substitutions based on the concept of the present invention.

Claims

1. A monitoring method, characterized in that: include: Acquiring a physiological signal, wherein the physiological signal includes a waveform, a parameter sequence, or a combination of the two; Obtain the correspondence between application scenarios and the working status of multiple artificial intelligence alarm sub-models; Determine the current application scenario based on user operations; According to the working status of the plurality of artificial intelligence alarm sub-models corresponding to the current application scenario, the corresponding artificial intelligence alarm sub-model is turned on or off; The acquired physiological signals are detected by the activated artificial intelligence alarm sub-models. Each artificial intelligence alarm sub-model uses the physiological signal as input to detect a type of abnormal situation and outputs an alarm message when an abnormal situation is detected. synthesizing the alarm information output by the activated artificial intelligence alarm sub-model; Output the integrated alarm information; in different application scenarios, the types of alarm information output are different.

2. The method according to claim 1, wherein The abnormal conditions to be detected by each artificial intelligence alarm sub-model are determined based on the length of the physiological signal required to detect the abnormal condition, and / or the preprocessing operation required to detect the abnormal condition, and / or the similarity of the waveforms corresponding to the abnormal condition.

3. The method according to claim 1, wherein The synthesizing of the alarm information output by the activated artificial intelligence alarm sub-model includes: The alarm information output by the enabled artificial intelligence alarm sub-model is synthesized according to the confidence level of the alarm information output by the enabled artificial intelligence alarm sub-model.

4. The method according to claim 1, wherein The method further comprises: Determine the priority of the alarm information output by the enabled artificial intelligence alarm sub-model.

5. The method according to claim 4, wherein The synthesizing of the alarm information output by the activated artificial intelligence alarm sub-model includes: The alarm information output by the enabled artificial intelligence alarm sub-model is synthesized according to the priority of the alarm information output by the enabled artificial intelligence alarm sub-model.

6. The method according to claim 3, wherein The combined structure of multiple artificial intelligence alarm sub-models is a serial structure, a parallel structure or a hybrid structure.

7. The method according to claim 6, wherein If the combined structure of the plurality of artificial intelligence alarm sub-models is a serial structure, then the alarm information output by the enabled artificial intelligence alarm sub-models is synthesized, including: The alarm information output by the activated artificial intelligence alarm sub-model, whose confidence is greater than the alarm confidence threshold and has the highest priority, is used as the integrated alarm information.

8. The method according to claim 6, wherein If the combined structure of the plurality of artificial intelligence alarm sub-models is a parallel structure, then the alarm information output by the enabled artificial intelligence alarm sub-models is synthesized, including: The alarm information with the highest confidence level and greater than the alarm confidence level threshold among the alarm information output by the activated artificial intelligence alarm sub-model is used as the integrated alarm information.

9. The method according to claim 3, wherein The method further comprises: Generate an alarm configuration interface, wherein the alarm configuration interface displays the working status of each artificial intelligence alarm sub-model, wherein the working status includes an enabled state and a disabled state; In response to the user's operation on the working status, the working status of the corresponding artificial intelligence alarm sub-model is updated.

10. The method according to claim 9, wherein The alarm configuration interface also displays the alarm confidence threshold of the artificial intelligence alarm sub-model that is in an enabled state; In response to the user's operation on the alarm confidence threshold, the alarm confidence threshold of the corresponding artificial intelligence alarm sub-model is updated.

11. The method according to claim 9, wherein The alarm configuration interface also displays a combination structure of multiple artificial intelligence alarm sub-models; In response to the user's operation on the combined structure, the combined structure of the plurality of artificial intelligence alarm sub-models is updated.

12. The method according to claim 1, wherein Before detecting the acquired physiological signals respectively by the activated artificial intelligence alarm sub-model, the method further includes: Analyzing the physiological signal to obtain a signal quality index of the physiological signal; The process of detecting physiological signals by multiple artificial intelligence alarm sub-models is controlled according to the signal quality index.

13. The method according to claim 12, wherein: The controlling of the process of detecting physiological signals by multiple artificial intelligence alarm sub-models according to the signal quality index includes: determining the working status of the plurality of artificial intelligence alarm sub-models according to the signal quality index; The acquired physiological signals are detected respectively through the artificial intelligence alarm sub-model in the enabled state.

14. The method according to claim 12, wherein: The controlling of the process of detecting physiological signals by multiple artificial intelligence alarm sub-models according to the signal quality index includes: The confidence level of the alarm information output by each artificial intelligence alarm sub-model is positively correlated with the signal quality index.

15. The method according to claim 12, wherein The analyzing the physiological signal to obtain the signal quality index of the physiological signal includes: A signal quality index of the physiological signal is determined according to at least one of an amplitude, a slope, and a power spectrum of the physiological signal.

16. The method according to claim 15, wherein Determining the signal quality index of the physiological signal according to at least one of the amplitude, slope, and power spectrum of the physiological signal includes: The signal quality index of the physiological signal is determined according to the following formula: δ=1-(α+β+2*γ) / 4; Wherein, δ represents the signal quality index, α represents the ratio of the amplitude of the physiological signal exceeding the preset amplitude range, β represents the ratio of the slope of the physiological signal exceeding the preset slope range, and γ represents the power ratio of the frequency of the physiological signal exceeding the preset frequency range.

17. The method according to claim 12, wherein The analyzing the physiological signal to obtain the signal quality index of the physiological signal includes: The physiological signal is input into a pre-trained signal quality assessment model to obtain a signal quality index of the physiological signal, wherein the signal quality assessment model is trained based on the physiological signal labeled with the signal quality index.

18. The method according to claim 12, wherein The method further comprises: Outputting a signal quality index of the physiological signal.

19. A monitoring device, characterized in that: include: A signal acquisition circuit, which acquires physiological signals using sensor accessories connected to the patient; Output module, used for outputting alarm information; Memory, used to store programs; A processor, configured to implement the monitoring method according to any one of claims 1 to 18 by executing the program stored in the memory.

20. A computer-readable storage medium, characterized in that The device comprises a program, which can be executed by a processor to implement the monitoring method according to any one of claims 1 to 18.

Images