AI eyewear medical risk grading cues and interrupt control method and system

Abstract

The application relates to the field of intelligent medical wearable devices and discloses an AI glasses medical risk grading prompt and interruption control method and system. Bedside images, voice, posture data and instrument readings are acquired, and after timestamp alignment, a standardized medical event containing a risk score, an emergency degree, a scene weight and a high-risk prompt lower limit is generated. A comprehensive prompt score is calculated and compared with the high-risk prompt lower limit to determine the final prompt level. According to the on-site state parameters, the sterile state and the state recognition confidence, the interruptable level is determined, and the interruption control instruction is generated in combination with the risk increment. The event is recorded, queued, weakly reminded or covered. While outputting the local prompt, an asynchronous evidence-based task is triggered, the clinical guideline is backfilled, and in the case of conflict, review is prompted. The doctor's feedback is collected to update the threshold and cooling parameters. The system can reduce the interference of invalid prompts on bedside operations, improve the timeliness of high-risk event reach and the traceability of disposal.

Description

Technical Field

[0001] This application relates to the field of intelligent medical wearable device technology, specifically to a method and system for AI glasses medical risk classification prompts and interruption control. Background Technology

[0002] With the development of smart wearable devices and hospital information systems, AI glasses are beginning to be applied in scenarios such as bedside diagnosis and treatment, vital sign monitoring, medical staff collaboration, and risk alerts. These devices can acquire patient status, doctor's procedures, and medical equipment readings through cameras, microphones, posture sensors, and data connections with bedside instruments. They can then alert medical staff to any abnormalities detected via screen, sound, or vibration. Compared to fixed monitoring terminals, AI glasses are characterized by being worn on the body, providing a close-up view of the situation, and offering faster alerts, making them suitable for auxiliary reminders during continuous bedside diagnosis and treatment.

[0003] Current wearable medical alert devices typically focus on identifying abnormal indicators or outputting a single alarm when providing risk warnings, rarely considering whether the doctor is currently conducting a consultation, physical examination, performing aseptic procedures, recording medical records, or handling an emergency. If all alerts pop up immediately, low-risk or repetitive alerts can easily occupy the doctor's vision, hearing, or attention, affecting the current medical procedure. If a simple silence or delay strategy is adopted to reduce interference, high-risk events such as critical values, rapidly deteriorating vital signs, and medication risks may not reach the on-site doctor in a timely manner. Therefore, current technology struggles to achieve a stable balance between reducing unnecessary disturbances and ensuring timely alerts for high-risk events.

[0004] Furthermore, bedside scenarios often involve multiple sources of risk events simultaneously, including monitors, ventilators, infusion pumps, image recognition, voice recognition, and electronic medical records. Existing systems rely heavily on fixed priorities or arrival order to handle these concurrent events, lacking a dynamic scheduling mechanism that considers risk level, urgency, waiting time, and source reliability. This can easily lead to prolonged delays in secondary but still critical events. While some systems can access medical guidelines or knowledge bases for reference, external evidence-based results are delayed. If a high-risk alert requires complete evidence before being output, it impacts the timing of bedside intervention. Relying solely on local model judgments lacks subsequent evidence backfilling and conflict verification mechanisms. Therefore, an AI-powered glasses-based medical risk grading and interruption control solution is needed, which integrates medical risk levels, physician interruptibility status, alert queue scheduling, and asynchronous evidence backfilling. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this application provides a method and system for AI glasses to provide medical risk classification prompts and interruption control. This solves the problems in existing technologies where wearable medical devices do not combine the doctor's current operating status, environmental limitations, and event priorities for prompt control, which can easily lead to low-value prompts interfering with medical operations, delays or omissions of high-risk prompts, queuing and delays for multiple concurrent events, single prompt channels, and insufficient evidence-based support.

[0006] To achieve the above objectives, this application provides the following technical solution: The first aspect of this application provides a method for AI glasses to provide medical risk grading prompts and interruption control, including the following steps: Acquire bedside multimodal data and instrument readings, align them according to timestamps, and encapsulate them into standardized medical events containing risk scores, urgency levels, scenario weights, and high-risk warning thresholds.

[0007] The weighted sum of risk score, urgency level and scenario weight is calculated to obtain a comprehensive warning score. The comprehensive warning score is used to map the basic warning level, and the larger value between the basic warning level and the lower limit of the high-risk warning is taken to determine the final warning level.

[0008] The field status parameters and the comprehensive confidence score of status identification are extracted from bedside multimodal data. The field status parameters include aseptic status parameters. Based on the mapping of the field status parameters to the basic breakable level, the basic breakable level is corrected using the comprehensive confidence score of status identification to obtain the temporary breakable level. The unauthorized triggering flag is calculated based on the incremental calculation of the actual risk score, and the final breakable level is determined by combining the temporary breakable level.

[0009] The system utilizes the final alert level and the final interruptible level to perform event triage for standardized medical events and generate interruption control commands. It calls the alert output module to execute the output of the physical display channel, triggers asynchronous evidence-based tasks to extract clinical guidelines, and backfills evidence-based data.

[0010] Furthermore, in the AI ​​glasses medical risk grading and interruption control method, the process of determining the final warning level includes: determining whether the standardized medical event hits a high-risk direct access condition or a continuous abnormal condition. When the high-risk direct access condition is hit, the lower limit of the high-risk warning is forcibly set to level R3; when the continuous abnormal condition is hit, the lower limit of the high-risk warning is forcibly set to level R2. The basic warning level is mapped to an integer constant in the memory structure, and the larger value between the integer constant and the forcibly set lower limit of the high-risk warning is taken as the final warning level.

[0011] Furthermore, in the AI ​​glasses medical risk grading and interruption control method, the process of correcting the basic interruptible level to obtain the temporary interruptible level includes: comparing the overall confidence level of state recognition with the state confidence threshold. When the overall confidence level of state recognition is less than the state confidence threshold, a degradation factor is introduced. The basic interruptible level is subtracted from the degradation factor to obtain a numerical result, and the maximum value between zero and this numerical result is determined as the temporary interruptible level.

[0012] Furthermore, in the AI ​​glasses medical risk grading and interruption control method, the process of calculating the unauthorized trigger flag and establishing the final interruptible level includes: extracting the new critical source flag of standardized medical events. When the value of the new critical source flag is equal to 1, or the actual risk score increment is greater than or equal to the risk increment threshold, or the value of the high-risk warning lower limit is equal to 3, the unauthorized trigger flag is set to the value of 1, which corresponds to allowing unauthorized access. The product of the unauthorized trigger flag and the constant 3 is compared with the temporary interruptible level, and the larger value is taken as the final interruptible level.

[0013] Furthermore, in the AI ​​glasses medical risk classification prompting and interruption control method, the process of executing event diversion and generating interruption control instructions includes: pushing standardized medical events into a waiting queue using the final prompt level and the final interruptible level; obtaining the queue waiting time, source reliability coefficient, and maximum tolerable waiting time scale of the standardized medical events in the waiting queue; calculating the ratio of the queue waiting time to the maximum tolerable waiting time scale, and introducing a constraint factor to constrain the upper limit of this ratio to not exceed 1. The final prompt level is numerically processed to obtain a final prompt level scalar value. Using different weight coefficients, a weighted sum is performed on the final prompt level scalar value, urgency, the above ratio, and the source reliability coefficient to obtain a dynamic dequeue priority. Interruption control instructions are generated sequentially from largest to smallest based on the dynamic dequeue priority.

[0014] Furthermore, in the AI ​​glasses medical risk grading prompt and interruption control method, the process of executing physical channel output and evidence-based backfilling includes: extracting sterility status parameters; when the sterility status parameter value is equal to 1 or the final interruptible level value is less than or equal to 1, blocking the bright occlusion pop-up and the regular buzzer alarm channel in the interruption control command, redirecting the physical display channel to call the bone conduction unit to execute bone conduction voice broadcast and simultaneously call the screen display unit to execute micro-frequency flicker at the edge of the field of view. Analyzing the recommended risk level associated with clinical guidelines, numerically processing the final prompt level to obtain a scalar value for the final prompt level, comparing the difference between the recommended risk level and the scalar value of the final prompt level; when the difference is greater than or equal to the conflict judgment threshold, establishing an abnormal conflict state; in the abnormal conflict state, maintaining the output of the already effective physical display channel unchanged and generating an additional graphic message carrying a review-required label.

[0015] This application separates the risk characteristics of medical data from the physical state of the operational site. The alert level is determined by quantitative indicators of the medical data combined with hard constraints, while the interruptibility level is determined by the operator's current action state and environmental conditions. In determining these levels, this solution compares the overall confidence level of the status identification with a threshold, introducing a subtraction factor to restrict interruption permissions. When a new critical source or risk mutation occurs, an overriding logic is triggered through numerical multiplication and extreme value selection, overriding the original environmental state restrictions. For event triage, a weighted summation model incorporating time ratios and constraint factors is constructed to calculate dynamic dequeue priorities. Furthermore, the generated logical levels are mapped to the channel switches of the underlying devices, executing the redirection of the physical display channel, and a conflict review branch is set when a difference exists between the local alert level scalar value and the risk level of external evidence-based guidance, reaching a threshold.

[0016] The second aspect of this application provides an AI glasses medical risk grading prompt and interruption control system, including an AI glasses terminal, an edge computing terminal, and an in-hospital server terminal connected in communication.

[0017] The AI ​​glasses are used to acquire bedside multimodal data and send it to the edge computing end; they also receive interruption control commands and call the prompt output module to execute the output of the physical display channel.

[0018] The edge computing terminal is used to receive bedside multimodal data and instrument readings, align them according to timestamps, and encapsulate them into standardized medical events containing risk scores, urgency levels, scenario weights, and high-risk warning thresholds. The weighted sum of risk scores, urgency levels, and scenario weights is calculated to obtain a comprehensive warning score. The comprehensive warning score is used to map a basic warning level, and the larger value between the basic warning level and the high-risk warning threshold is taken to determine the final warning level.

[0019] The edge computing terminal is also used to extract on-site state parameters and the comprehensive confidence level of state recognition. The on-site state parameters include aseptic state parameters. Based on the on-site state parameters, a basic interruptible level is mapped. The basic interruptible level is then corrected using the comprehensive confidence level of state recognition to obtain a temporary interruptible level. An overstepping trigger flag is calculated based on the actual risk score increment. Combined with the temporary interruptible level, the final interruptible level is established. The edge computing terminal is also used to perform event triage on standardized medical events using the final alert level and the final interruptible level, and to generate interruption control commands that are sent to the AI ​​glasses terminal, triggering asynchronous evidence-based tasks to the hospital server.

[0020] The hospital's server is used to receive asynchronous evidence-based tasks, extract clinical guidelines, and backfill evidence-based data to the edge computing end.

[0021] Furthermore, in the AI ​​glasses medical risk grading prompt and interruption control system, the edge computing end is specifically used to extract aseptic state parameters. When the aseptic state parameter value is equal to 1 or the final interruptibility level value is less than or equal to 1, the interruption control command blocks the brightly lit pop-up window and the conventional buzzer alarm channel, redirecting the physical display channel to call the bone conduction unit to execute bone conduction voice broadcast and simultaneously call the screen display unit to execute micro-frequency flashing at the edge of the field of view. The AI ​​glasses end is specifically used to execute the corresponding output actions through the built-in bone conduction unit and screen display unit according to the received redirected interruption control command.

[0022] This application provides a method and system for AI glasses medical risk grading prompts and interruption control. It has the following beneficial effects: 1. This application combines the risk parameters of a medical event with the operational status by separately calculating the final alert level and the final interruptibility level. When the overall confidence level of the status identification falls below a threshold, a downgrade factor is introduced to restrict interruption permissions. Furthermore, when a new critical source is extracted or the actual risk score increment reaches a set condition, an overreach trigger flag is used to re-raise the interruptibility level. This setting maintains the continuity of medical staff's current operations while ensuring that high-risk medical events are promptly alerted, reducing the probability of operational errors due to equipment interference or risk oversights due to silence.

[0023] 2. In the event triage control, this application calculates the ratio of the queue waiting time of standardized medical events to the maximum tolerable waiting time scale, and combines this ratio with the final alert level scalar value, urgency, and source reliability coefficient to obtain a dynamic dequeue priority. This mechanism uses a constraint factor to limit the upper limit of the time ratio, allowing the priority of secondary priority events with longer waiting times to increase over time. This avoids a single high-risk event sequence from continuously occupying the system alert channel, thus achieving reasonable sorting and scheduling of concurrent medical events.

[0024] 3. When executing physical channel output, this application sets a feature to block the conventional buzzer and brightly highlighted pop-up window when the aseptic state parameter value is equal to 1 or the final interruptibility level value is too low. Instead, the prompt is redirected to bone conduction voice broadcast and micro-frequency flashing at the edge of the field of view. Simultaneously, the suggested risk level from clinical guidelines is extracted and compared with the local final prompt level scalar value. When the difference reaches a threshold, a graphic message with a "requires review" label is generated. These features reduce the reliance of audio-visual alarms on the operator's subjective vision and hearing, and by introducing external evidence-based data to provide a comparison and review pathway, the risk of false alarms from a single algorithm is reduced. Attached Figure Description

[0025] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0026] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0027] One or more embodiments are illustrated by way of example with reference numerals in the accompanying drawings. These illustrations do not constitute a limitation on the embodiments. Elements with the same reference numerals in the drawings are denoted as similar elements. Unless otherwise stated, the figures in the drawings are not to be limited by scale.

[0028] Figure 1 This is a diagram illustrating the architecture of the risk classification, warning, and interruption control system according to an embodiment of this application. Figure 2 This is a flowchart of the prompt interruption control method according to an embodiment of this application; Figure 3 This is a comparison chart showing the evolution of the adaptive threshold learning mechanism effect in the embodiments of this application. Detailed Implementation

[0029] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0030] The following disclosure provides numerous different embodiments or examples for implementing various structures of this application. To simplify the disclosure, specific examples of components and arrangements are described below. These are merely examples and are not intended to limit the scope of this application. Furthermore, reference numerals and / or letters may be repeated in different examples. Such repetition is for simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or arrangements discussed.

[0031] For ease of description, spatial relative terms may be used in the text to describe the relative position or movement of one element or feature relative to another element or feature, as shown in the figure. These relative terms include, for example, "inside," "outside," "middle," "outer," "below," "below," "above," "front," "back," etc. Such spatial relative terms are intended to include different orientations of the device in use or operation, other than those depicted in the figure. For example, if the device in the figure undergoes a positional flip, orientation change, or change of motion, these directional indications will change accordingly. For instance, an element described as "below other elements or features" or "below other elements or features" will subsequently be oriented "above other elements or features" or "above other elements or features." Therefore, the example term "below" can include both upper and lower orientations. The device may be otherwise oriented (rotated 90 degrees or in other directions), and the spatial relative descriptors used in the text will be interpreted accordingly.

[0032] Reference Figure 1 This application provides an AI glasses medical risk classification prompt and interruption control system, which may include: AI glasses terminal, edge computing terminal and hospital server terminal.

[0033] The AI ​​glasses are equipped with a visual acquisition module, a voice acquisition module, a posture recognition module, and a prompt output module. The prompt output module includes a screen display unit, a bone conduction unit, and a vibration unit. The AI ​​glasses acquire image data, voice commands, and posture data of the bedside environment, and send the raw medical events and candidate values ​​of the doctor's status to the edge computing end. The AI ​​glasses receive interruption control commands and output prompt information through the prompt output module.

[0034] The edge computing terminal is equipped with an event normalization module, a risk calculation module, a deduplication and merging module, and a prompt queue module. It receives data from the AI ​​glasses and readings from bedside instruments, extracting risk features. Based on internal rules, the edge computing terminal generates medical events with risk attributes, queue scheduling items, and interruption control commands.

[0035] The hospital's server-side deployment includes an asynchronous evidence-based module, an audit storage module, and a feedback learning module. The server-side receives anonymized event summaries, feedback records, and evidence-based return results transmitted from the edge computing terminal, and outputs the evidence-based basis, threshold configuration parameters, and audit log files.

[0036] Data flows between various terminals and modules via an event bus. The data flow process is associated according to session identifiers and event identifiers. The system generates a session identifier when a diagnosis / treatment session starts, and all subsequent generated raw medical events are assigned corresponding event identifiers. In the event of a network outage, the AI ​​glasses and edge computing terminals cache the data locally according to the data generation sequence number. After the network is restored, the data is retransmitted sequentially according to the sequence number, enabling data retention and traceability of medical events from acquisition, prompting and control, evidence-based processing to feedback learning.

[0037] Reference Figure 2 Based on the above system architecture, this application provides a method for AI glasses medical risk classification prompts and interruption control, including the following steps: S10: Acquire bedside multimodal data and generate standardized medical events. Specifically, the system acquires bedside image and voice data through a visual acquisition module and a voice acquisition module. Combined with bedside instrument readings, the event normalization module converts risk signals into standardized medical events. Standardized medical events include event identifier, source type, risk type, risk score, urgency, scenario weight, model confidence level, and high-risk warning lower limit.

[0038] S20: Extract risk features and calculate a comprehensive alert score to determine the final alert level. Specifically, the risk calculation module extracts risk features from standardized medical events and calculates a comprehensive alert score. Model confidence parameters and state parameters do not participate in the weighted calculation of the comprehensive alert score. Based on the comprehensive alert score and the high-risk direct access condition, the risk calculation module determines the basic alert level and the lower limit of the high-risk alert for the medical event, and takes the higher value between the two as the final alert level.

[0039] S30: Identify the current operation status and determine the interruptibility level. Specifically, the posture recognition module infers the doctor's current diagnosis and treatment stage and operation status based on the collected posture data, gaze data, and voice data. Combining the status priority determination rules, the current interruptibility level is determined and attached to the corresponding standardized medical event.

[0040] S40, generate interruption control instructions and execute prompt queue scheduling; specifically, the prompt queue module generates interruption control instructions based on the final prompt level and the interruptibility level. The deduplication and merging module compares the frequency of similar events based on the cooldown key value to determine whether to transfer the current prompt information to the background record, add it to the waiting queue, perform a weak reminder operation, or trigger a strong interruption operation that overwrites the current screen.

[0041] S50 outputs local prompts and triggers asynchronous evidence-based tasks. Specifically, when an immediate prompt action is deemed necessary, the AI ​​glasses display local safety prompts via the prompt output module. For standardized medical events of medium to high risk, the hospital server triggers asynchronous evidence-based tasks. The asynchronous evidence-based module retrieves medical guidelines and treatment recommendations, obtains the evidence-based results, and then populates the system interface with the evidence-based data.

[0042] S60 collects interactive feedback data and updates the scoring threshold parameters; specifically, the feedback learning module collects data on the adoption, ignoring, or delay of prompt actions, and updates the scoring threshold and cooldown time parameters within the system. Standardized medical events that hit the lower limit rule for high-risk prompts are not included in the frequency reduction and downgrade calculation.

[0043] In this embodiment, the specific implementation method of step S10, acquiring bedside multimodal data and generating standardized medical events, is described in detail in conjunction with the overall workflow. To achieve a comprehensive understanding of the medical scene, this step may specifically include the following sub-steps: S110 acquires multimodal data from the bedside and readings from bedside medical instruments. Specifically, the system uses a vision acquisition module deployed on the AI ​​glasses to acquire image data of the bedside environment. This image data typically includes the patient's facial features, surgical incision details, and the operational status of medical instruments. Simultaneously, the voice acquisition module collects ambient sound signals for later extraction of patient groans, instrument alarms, and specific commands issued by the doctor. To capture the doctor's posture, the posture recognition module continuously collects the doctor's head pitch and yaw angle data using inertial sensors. Meanwhile, the edge computing terminal receives real-time readings from bedside medical instruments via the hospital's local area network or near-field communication interface.

[0044] In one preferred embodiment, the bedside medical instrument includes a monitor, a ventilator, and an infusion pump. The readings cover heart rate, blood oxygen saturation, blood pressure, and equipment operating status parameters. The hardware drivers and underlying data reading logic for the image sensor, microphone array, and inertial sensor can be implemented using conventional embedded hardware-software co-processing solutions, which are well-known technologies in the field and will not be elaborated upon here.

[0045] S120: Timestamp alignment and fusion of multi-source heterogeneous data. Since visual data, voice data, posture data, and instrument readings have different sampling frequencies and transmission delays, direct overlay analysis can easily lead to misjudgments in state identification. Therefore, the event normalization module aligns various data types within a preset time window according to the absolute timestamp carried by the data packets. In this embodiment, the size of the preset time window can be dynamically set according to the sampling frequency of the specific bedside device and the network latency, typically set to the range of several hundred milliseconds to several seconds. The timestamp-aligned multimodal data constitutes a snapshot data set representing the current bedside diagnosis and treatment status, thereby providing a time-consistent input source for subsequent risk feature extraction.

[0046] S130 converts the aligned data into standardized medical events with a unified data structure. For the snapshot data set constructed above, the event normalization module performs anomaly detection and classification analysis. When physiological indicators exceed limits, visual abnormalities occur, or specific voice keywords are triggered, these are uniformly encapsulated as standardized medical events.

[0047] For identifying visual anomalies, the system employs a deep learning-based image classification model. This model utilizes a residual network structure, which sequentially includes convolutional layers, pooling layers, residual blocks, and fully connected layers. During processing, the input data is a three-channel image matrix that has undergone preprocessing such as scaling and cropping. The data flows through convolutional feature extraction and dimensionality reduction, and the fully connected layers output probability vectors corresponding to different visual risk categories (such as incision bleeding, tube detachment, etc.).

[0048] Before deployment, the image classification model undergoes supervised training using clinically abnormal scene image samples annotated by medical experts. The training process updates the network layer weights using a cross-entropy loss function. Similarly, for the extraction of specific speech keywords, the system utilizes an acoustic feature extraction component to convert the audio waveform into Mel-frequency cepstral coefficient features, which are then input into a pre-trained Hidden Markov Model or Recurrent Neural Network to output a text sequence that matches a pre-defined medical risk lexicon.

[0049] After initial screening and identification, the system generates a standardized medical event with a uniform structure for each identified anomaly. The standardized medical event includes fields such as event identifier, source type, risk type, risk score, urgency, scenario weight, model confidence, and high-risk warning lower limit.

[0050] Specifically, the event identifier is a unique sequence number assigned by the system, used to track the event throughout the entire data flow lifecycle. The source type indicates the input channel for the risk signal, and its sub-characteristics can be categorized as visual images, audio, gestures, or instrument interfaces. The risk type is used to define the specific medical attributes of the abnormality, and its sub-characteristics can be classified as abnormal vital signs, operational violations, or equipment malfunctions.

[0051] Risk score is a quantitative value characterizing the degree to which the current abnormal state deviates from the normal physiological baseline or operational norms. Urgency is a quantitative value characterizing the rate of risk deterioration or the urgency of the time window. In this embodiment, the event normalization module calculates the risk score based on the absolute difference between instrument readings and safety thresholds, and calculates the urgency based on the slope of the reading change over a unit of time. The system uses a minimax algorithm to uniformly map the risk score and urgency to a numerical range of 0 to 1. The upper and lower bounds on which this minimax algorithm relies are pre-calibrated based on the clinical survival limits of various physiological indicators or the physical range of medical equipment. This mapping process allows heterogeneous physical units to be compared and calculated on the same data scale.

[0052] Scenario weight is an adjustment coefficient set by the system based on the current department's attributes or the patient's specific underlying disease. For example, in an intensive care unit scenario, the scenario weight is set higher than in a general ward scenario for the same range of blood pressure fluctuations, enabling standardized medical events to reflect the risk tendency under specific medical conditions.

[0053] Model confidence is the certain probability that an algorithm model will output a classification and score for a given event. Model confidence reflects whether there is interference or ambiguity in the current perceived data, and its value ranges from 0 to 1.

[0054] The high-risk alert lower limit is the minimum level of protection set by the system for an event based on preset high-risk pass rules. These preset high-risk pass rules typically include, but are not limited to, clearly defined critical value indicators such as cardiac arrest and blood oxygen saturation dropping below a warning threshold. When specific risk characteristics that meet the criteria for life-threatening events or requiring immediate intervention are detected, the system forcibly writes a fixed high-risk alert lower limit value into the event's data structure. This completes the marking of underlying risks during the data encapsulation phase, reducing the risk of lowering the event's processing priority due to external interference in subsequent processing logic.

[0055] In this embodiment, the specific implementation method for extracting risk features and calculating a comprehensive warning score in step S20 to determine the final warning level, based on the standardized medical event data structure, is described in detail. This step aims to quantify the risk level of an event through a numerical model, and specifically includes the following sub-steps: S210: Extract risk characteristic parameters of standardized medical events and calculate a comprehensive alert score. The risk calculation module within the edge computing terminal analyzes the received standardized medical events, extracting the risk score, urgency, and scenario weight. To avoid local misjudgments of abnormal states caused by a single parameter, the system uses preset weighting factors to weight and fuse the above multi-dimensional parameters. Let the comprehensive alert score be... The risk score is The urgency level is Scene weight is The corresponding weighting factors are denoted as follows: , as well as Comprehensive hint score The calculation formula is as follows: ; In this formula, each risk characteristic parameter , , The values ​​have all been mapped to the range of 0 to 1 during the preprocessing stage. Meanwhile, the preset weighting factors... , , The sum of the values ​​equals 1, thus ensuring the overall prompt score. The value range is limited to 0 to 1. As a preferred approach, the specific value of the weighting factor is pre-configured by the administrator based on the departmental environment in which the system is deployed, to adapt to the sensitivity tendencies of various risk characteristics in different clinical scenarios. For example, in the intensive care unit, the weighting factor corresponding to urgency can be appropriately increased. To enhance the sensitivity of responses to disease mutation indicators.

[0056] S220 maps basic prompt levels based on the comprehensive prompt score and performs independent, decoupled calculations of perception parameters. After obtaining the comprehensive prompt score... Subsequently, the risk calculation module maps these to basic warning levels from low to high. In this embodiment, the basic warning levels are divided into R0, R1, R2, and R3 levels, respectively. To ensure the validity of the underlying numerical operations of the computer program, the system maps the above labels to the integer constants 0, 1, 2, and 3 in the memory structure, respectively.

[0057] The system presets three dynamically adjustable segmented thresholds, denoted as the first initial threshold. Second initial threshold and the third initial threshold As a preferred initial configuration, these three thresholds can be set to 0.25, 0.50, and 0.75 by default. The specific mapping rule is set as follows: when the overall prompt score... Less than When, it is judged as level R0; when Greater than or equal to and less than When, it is judged as R1 level; when Greater than or equal to and less than When, it is judged as R2 level; when Greater than or equal to At that time, it was determined to be level R3.

[0058] In determining the basic alert level, the system establishes a parameter decoupling control mechanism. The model confidence parameters and doctor status parameters carried in standardized medical events do not participate in the overall alert score. The weighted calculation ensures the objectivity and authenticity of risk assessment at the underlying logic level.

[0059] S230 determines the lower limit of the high-risk warning based on preset conditions and calculates the final warning level. The risk calculation module determines whether the current standardized medical event meets the high-risk direct access condition or the continuous abnormal condition. When the event meets the high-risk direct access condition (such as a decrease in blood oxygen saturation exceeding 5%), the system forcibly sets the lower limit of the high-risk warning to R3 level (numerical mapping 3). When the continuous abnormal condition is met, the system forcibly sets the lower limit of the high-risk warning to R2 level (numerical mapping 2). If neither condition is met, the default level is R0 (numerical mapping 0).

[0060] After completing the above matching conditions, the system compares the basic alert level with the lower limit of the high-risk alert. Let the final alert level be... The lower limit for high-risk warnings is Based on the comprehensive prompt score The base hint level obtained from the mapping is denoted as Final warning level The decision logic follows the principle of taking the larger of the two values, as expressed below: ; This mechanism ensures that when the system detects clear high-risk characteristics, the final alert level for the event can avoid falling below the lower limit of high-risk alerts. Only when the base alert level calculated by the formula is higher than or equal to the lower limit of the high-risk alert is the original base level output. This ensures the underlying risk response rate while providing a safe numerical benchmark for the system's subsequent alert queue scheduling and overlay logic.

[0061] In this embodiment, the specific implementation of step S30, identifying the current operational state and determining the interruptibility level, is described in detail, combining the data structure of standardized medical events and the risk level calculation results. This step aims to establish a rigorous mathematical evaluation model using multimodal sensor data to quantitatively assess the degree to which the doctor's current attention is occupied, providing a decision-making basis for subsequent anti-interference control measures. Specifically, it may include the following sub-steps: S310: Extract multi-dimensional state features and construct a set of on-site state parameters. The system extracts multi-dimensional features of the doctor's real-time dynamics by receiving head pitch angle data, yaw angle data, gaze tracking coordinates, and microphone audio stream transmitted from the AI ​​glasses. For the mapping of gaze tracking data points and the detection of speech activity in the audio stream, those skilled in the art can use conventional pupil-corneal reflection algorithms and energy threshold determination methods, which are well-known technologies in the field and will not be elaborated upon here.

[0062] To accurately identify the doctor's current hand movements and whether a sterile environment is being used, the system employs a deep learning-based object detection model. As a preferred approach, this model utilizes a feature pyramid network structure, which includes a convolutional backbone for basic feature extraction, a neck network for multi-scale feature fusion, and a detection head module for outputting predicted bounding boxes. In actual operation, the input data is an image matrix containing the doctor's perspective, output by the AI ​​glasses' visual acquisition module. This image matrix undergoes preprocessing before being input into the model, specifically scaling and cropping the images to a specific spatial resolution (e.g., 640×640 pixels) and normalizing them to an RGB three-channel format. After forward propagation, the detection head module outputs the target bounding boxes and their corresponding classification confidence scores. Before deployment, the object detection model undergoes supervised training using a historical image dataset collected from real-world clinical scenes and annotated by professionals. During training, classification cross-entropy loss and bounding box regression loss functions are used to update network weights.

[0063] After acquiring the aforementioned basic sensing data, the system establishes a set of on-site state parameters using a preset rule engine. Let the parameters for the diagnosis and treatment phase be... The aseptic state parameters are: The overall confidence level for state recognition is Specifically, the rule engine performs logical mapping based on feature cross-determination: when the gaze remains fixed on the patient's face and the microphone captures continuous conversational voiceprint features, a decision is made... During the consultation phase; when the inertial sensor shows frequent head tilting and the target detection model fails to detect surgical instruments, a decision is made. During the physical examination or recording phase; when the target detection model continuously outputs bounding boxes of someone holding a specific surgical instrument and those bounding boxes are located in the patient's wound area, a determination is made. This is the bedside procedure stage. Simultaneously, the system determines the aseptic status of the current procedure based on the output sterile glove label; when in an aseptic procedure state... When in a non-sterile or unknown state .

[0064] S320 constructs a basic interruptibility level mapping function and performs conservative degradation calculations. After obtaining the set of on-site state parameters, the system converts the attention occupancy state into a quantified basic interruptibility level according to internally set priority rules. Let the basic interruptibility level be... The system establishes mapping functions. Its segmentation logic is as follows: ; In this formula, Represents an uninterruptible state. Characterizes a low interruptibility state. The characterization can be used to weakly indicate a state. The representation can strongly remind you of the status.

[0065] Furthermore, considering the environmental noise interference during multimodal sensor fusion and inference, the system introduces a conservative degradation mechanism. Let the degradation factor be... The preset state confidence threshold is Degradation factor The value is determined by the overall confidence level of state recognition. Decision: When hour ,when hour Combining basic abilities allows for interruption of certain levels. The system calculates and obtains the temporary interruptible level after the downgrade correction. The calculation formula is as follows: ; By introducing the maximum value function The system can effectively avoid the logical dead zone caused by excessive degradation when the base level is already 0, resulting in negative values. As a specific implementation method, this preset state confidence threshold... It is usually set between 0.6 and 0.7.

[0066] S330, incorporating the overreach triggering mechanism, calculates the final interruptibility level and completes the state attachment. In high-pressure clinical environments, the low interruptibility level under normal circumstances (i.e., the temporary interruptibility level after downgrade correction) is... The system will silently enqueue standardized medical events, suppressing routine alert output. To prevent this control logic from causing underreporting of serious medical risks, an unauthorized trigger judgment logic has been constructed.

[0067] Set the unauthorized access trigger flag as Where 1 indicates that unauthorized access is permitted, and 0 indicates that unauthorized access is prohibited. The actual risk score increment is set to... The preset risk increment threshold is Whether the current medical event is a newly emerging critical source is identified as follows: And the lower limit of high-risk warning is Unauthorized access trigger flag The determination equation is as follows: ; Among them, the logical OR operator This indicates that exceeding authority can be triggered if any one of the conditions is met. Preset risk increment threshold. The threshold can be set to 0.15 to 0.20 based on specific physiological indicators. After completing the above overreach determination, the system calculates the final interruptible level. Ultimately, the interruptible level... The comprehensive calculation formula is expressed as follows: ; This formula ensures that, under normal conditions, the system strictly adheres to the temporary interruptible level after the degradation correction. Run; however, once an emergency high-risk situation is determined, i.e., the unauthorized trigger flag is equal to 1 ( When this occurs, the system will ignore environmental constraints and directly force the final interruptible level to the highest level 3, thereby allowing the underlying layer to execute mandatory anti-interference coverage prompts.

[0068] Ultimately, the system will determine the final interruptible level. and aseptic status parameters As an independent attribute field, it is added back into the data structure of the corresponding standardized medical event, thereby integrating risk attributes and on-site environmental constraints into the underlying data object, providing a complete input basis for generating differentiated interruption control instructions in the next step.

[0069] In this embodiment, the specific implementation of step S40, which generates interruption control instructions and executes the prompt queue scheduling, is described in detail, taking into account the determined final prompt level and the final interruptibility level. This step aims to address the problem of information overload in high-concurrency clinical environments by achieving a balance between interference suppression and crisis communication through a dynamic scheduling mechanism. Specifically, it may include the following sub-steps: S410, Perform similar event judgment and deduplication cooling operation. After receiving a standardized medical event with additional status parameters, the prompt queue module in the edge computing terminal extracts the historical sequence of similar events based on the key identifiers carried by the event. In this embodiment, the judgment criteria for similar events are that the patient identifier, source type, and risk type are completely identical. To avoid frequent alarms caused by minor fluctuations in instrument readings, the system sets a cooling-off period for similar events. Specifically, the system presets a normal cooling-off time of [missing information]. For high-risk events, the preset high-risk re-reporting interval is 100%. ,in The value is strictly less than As a preferred parameter configuration, It can be set to 3 to 5 minutes, and It can be set to 30 to 60 seconds to ensure high-frequency monitoring in critical situations.

[0070] Similar events within the cooling-off period are by default intercepted by the system and converted into background log merging operations. To prevent serious medical risks from being incorrectly masked by the cooling-off mechanism, the system introduces an icebreaking verification based on risk increments. The actual risk score increment is set to... The preset risk increment threshold is When the actual risk score increment for the same risk event meets the following conditions... If the current standardized medical event includes a new critical source (for example, in addition to the existing heart rate abnormality alarm, the same patient has new characteristic data of rapid blood oxygen drop), the system will immediately release the current cooling constraint and allow the event to enter the subsequent processing node.

[0071] The S420 uses state-aware event routing and queue sorting logic. Valid events that have passed cooling and filtering enter the dynamic routing node. The system determines the routing based on the event's final notification level. The current doctor's final interruptible level Implement the traffic splitting strategy. When When the system reaches level R0, it does not generate front-end display commands but directly writes them to the back-end storage log. Equal to R1 level and current doctor When the value is less than or equal to 2, and when Equal to R2 level and current doctor When the value is less than or equal to 1, it means that the doctor is currently in a busy state of physical examination, consultation, or higher-level (such as aseptic operation). The system pushes this event into the waiting queue, suspending the display command until the multimodal sensor determines that the doctor's status has entered an idle window or the current treatment round has ended. In addition, when When it reaches level R3, the system ignores environmental constraints and directly triggers subsequent dequeueing and interruption displays without entering the regular waiting queue, but retains its mapping scalar to be compatible with the system's underlying unauthorized tracing mechanism.

[0072] For multiple events stuck in the waiting queue, the system constructs a dynamic dequeue priority calculation model. Let the dynamic dequeue priority be... The final scalar value of the prompt level is... The urgency level is The queue waiting time is The preset maximum tolerable waiting time scale is The source reliability coefficient is The weighting coefficients of the priority parameters are denoted as follows: , , as well as In this model, The specific mapping rule is as follows: R1 level corresponds to the value 1, R2 level corresponds to the value 2, and R3 level corresponds to the value 3. Dynamic dequeue priority. The calculation formula is as follows: ; By introducing a maximum tolerance waiting time scale Queue waiting time By performing division operations and dividing the level scalar by the maximum constant 3, the system strictly normalizes all heterogeneous variables to a scale between 0 and 1. At this point, the weighting coefficients must satisfy the normalization condition. Meanwhile, to prevent excessively long waiting times from causing low-risk events to abnormally outpace newly arrived high-risk events, the system sets a limiting factor to... The ratio cannot exceed 1. This is achieved by introducing a queue waiting time. The system ensures that the priority of low-level prompts in the queue increases linearly over time, preventing queue starvation. (Source reliability coefficient) These are fixed parameters set by the system based on the signal source channel. For example, the reliability coefficient of numerical alarms directly output by a vital signs monitor is higher than that of environmental voice alarms extracted by a microphone. Within the queue, dequeue priority is always dynamically calculated. The commands are sorted out in descending order and the corresponding interrupt control commands are generated sequentially.

[0073] S430 generates interruption control commands and implements multi-channel adaptive display. For events that do not require queuing or have reached the dequeue condition, the system generates a specific physical display action allocation scheme based on the final prompt level. When When the system reaches R2 level, it generates a weak reminder control command, calls the micro-interaction interface on the AI ​​glasses, displays a miniature status icon in the corner area of ​​the retinal projection screen, and is accompanied by a short-term low-frequency vibration inside the temple.

[0074] when When the event reaches level R3, the system initiates a forced interruption and overwrite mechanism. For events at this level, the system forcibly writes an overwrite flag and sets its parameter value to true when generating instructions. The edge computing terminal sends an instruction carrying this flag to the AI ​​glasses. Upon receiving the instruction, the AI ​​glasses immediately suspend or clear the current R0 to R2 level prompts on the screen and exclusively occupy the display channel.

[0075] During the execution of R3 level forced interruption, the system calls the aseptic state parameters calculated in the early stage. With the final interruptible level Perform final channel verification. When determined... or At this time, it indicates that the doctor is currently performing a critical aseptic or critical procedure. To minimize the probability of operational errors caused by audio-visual stimulation, the system automatically blocks the high-brightness pop-up window and the conventional buzzer alarm channel, redirecting the physical display channel to bone conduction voice broadcast combined with micro-frequency flashing at the edge of the field of view. The bone conduction channel can bypass external environmental noise to transmit critical values ​​to the inner ear, and the edge micro-frequency flashing can provide visual warnings without obstructing the central area of ​​the surgical field of view. The audio stream input to the bone conduction driver chip and the area refresh rate control of the optical projection module can be implemented using conventional low-level hardware driver development techniques, which are well-known technologies in the field and will not be elaborated upon here. After the above adaptive channel selection, the system outputs the final interruption control command to the glasses execution end, completing the closed-loop warning for this medical event.

[0076] In this embodiment, the specific implementation of step S50, which outputs local prompt information and triggers asynchronous evidence-based tasks, is described in detail, based on the interruption control instructions generated earlier. This step establishes a collaborative operating mechanism that prioritizes local safety prompts and coordinates asynchronous evidence-based backfilling, aiming to balance the requirements of immediate bedside response to high-risk events with the rigor of subsequent clinical decision-making. Specifically, it may include the following sub-steps: The S510 performs local real-time prompts and asynchronous evidence-based triggering judgments. For standardized medical events judged as R2 or R3, after the edge computing terminal issues a command, the AI ​​glasses immediately output physical prompts through the corresponding hardware channel. To provide more detailed medical evidence for on-site operations, the system synchronously establishes asynchronous evidence-based triggering logic. Let the previously acquired model confidence level be... The preset model confidence threshold is The parameter indicating whether a medical event includes medication recommendations is: As one specific implementation method, the preset model confidence threshold is... It can be set between 0.75 and 0.85. The specific triggering rules are set as follows: for R0 and R1 level events, external resource calls are not triggered by default; for R2 level events, when the conditions are met... or When an event occurs, an asynchronous evidence-based task is triggered. For R3 level events, the system forcibly generates an asynchronous evidence-based task regardless of other conditions. Under this mechanism, the output of local security alerts and the asynchronous evidence-based requests sent to the hospital's server occur in parallel. The local alert action never waits for the return of external network evidence, thus avoiding the risk of clinical delays due to network latency at the underlying logic level.

[0077] S520 performs asynchronous evidence-based retrieval and handles abnormal state boundaries. Upon receiving an evidence-based task request carrying a unique event identifier, the hospital server retrieves matching clinical guidelines and treatment recommendations from its internally deployed medical knowledge base. For this retrieval process, those skilled in the art can employ text-vectorized cosine similarity matching algorithms or conventional relational database query techniques, which are well-known technologies in the field and will not be elaborated upon here. To address abnormal boundary situations during the retrieval interaction process, the system establishes a clear fault-tolerance defense. When the evidence-based task takes longer than a preset timeout threshold or the server returns a failure status code, the system determines the task to have timed out or failed. In this case, the system retains the triggered local security alert and marks the audit anomaly field in the background data table.

[0078] When the retrieval is successful and external evidence is successfully returned, the system performs a consistency check between the locally calculated risk level and the external guideline recommendations. To avoid ambiguity in the judgment criteria, this consistency check is implemented as follows: the system extracts the suggested risk level associated with the external evidence text and compares it with the local final prompt level scalar value. If the difference between the two levels is greater than or equal to a preset conflict judgment threshold (e.g., spanning two levels, locally set to R3 while the external guideline recommendation is only R1), the system sets the return status to conflict indicator. In the conflict state, the system does not automatically revoke the effective local prompt, nor does it directly interfere with the doctor's on-site judgment. Instead, it generates an additional graphic message with a "requires review" label. This graphic message is pushed to the secondary display layer of the AI ​​glasses or the associated tablet device for manual confirmation by the clinician.

[0079] S530, a comprehensive application example combining real-world clinical data stream execution. To clearly demonstrate the collaborative operation of various threshold parameters and logical branches, a typical bedside case of sudden change in patient condition is used as an example. Multimodal sensors capture a rapid drop in the patient's blood oxygen saturation from 94% to 88%, while microphone audio stream and gaze data infer that the doctor is currently in the silent auscultation examination phase. The system extracts various dimensional parameters of this event: risk score... The urgency level is Scene weight is The system is based on preset weighting factors. , , Calculate the overall prompt score The calculation process is as follows: ; The overall score is shown in this calculation result. Upon reaching 0.84, exceeding the preset threshold of 0.75, the basic alert level is mapped to R3. Simultaneously, the rapid drop in blood oxygen triggers a preset high-risk direct access condition, and the system forcibly sets a lower limit for the high-risk alert. The level is R3. Based on the logic of taking the larger value, the final suggested level is... It has been designated as R3 level.

[0080] During the control command generation phase, considering the constraints of the doctor's physical examination, the system activates the coverage flag and issues a physical display command consisting of bone conduction voice broadcasting combined with micro-frequency flashing at the edge of the field of view. Simultaneously, the backend asynchronously generates an evidence-based retrieval task based on R3-level forced triggering rules, specifically retrieving emergency procedures for hypoxemia. Once the retrieved evidence is returned, it is appended to the underlying data structure of the original medical event using the same event identifier. After the doctor completes the on-site emergency treatment, they can review the supplementary evidence through voice interaction and generate a feedback record indicating adoption or disregard, thus providing comprehensive data support for risk alerts, action execution, evidence tracing, and audit records within a single data lifecycle.

[0081] In this embodiment, the specific implementation of step S60, which involves collecting interactive feedback data and updating the scoring threshold parameters, is described in detail. This step aims to enable the system to adaptively optimize the sensitivity of the control strategy during continuous operation by introducing a long-term closed-loop learning mechanism, while simultaneously safeguarding the safety baseline of medical risks through rigid rules. Specifically, it may include the following sub-steps: The S610 collects multi-dimensional terminal feedback data and calculates statistical ratios. The edge computing end and the AI ​​glasses work collaboratively to capture doctors' behavioral feedback in response to various prompts in real time. To facilitate system recognition and quantification, terminal feedback operations are standardized into four distinct enumerated states: Accept, Ignore, Delay, and Modify. Specifically, when a doctor clicks the confirmation button on the glasses or issues a corresponding treatment order in the system, the system marks it as accepted; when a doctor directly slides to close the prompt window or the prompt remains for more than a set time without any interaction, the system marks it as ignored. The hospital server persistently stores the aforementioned behavioral state data in a historical database according to department attributes, doctor roles, scenario types, and risk types.

[0082] To enable quantitative analysis, the system sets a fixed statistical time window. For a specific type of standardized medical event, the system counts the total number of times it occurs within that time window. The number of times an item is marked as adopted is denoted as . The number of times a character is marked as ignored is denoted as . Based on this, the system calculates the acceptance rate for this type of event through numerical calculations. With neglect rate The relevant calculation formulas are as follows: ; ; S620 executes a closed-loop threshold learning algorithm based on statistical features. When the statistical features indicated at a certain level meet the update conditions, the system calls the aforementioned initial segmentation threshold (i.e., the first initial threshold) corresponding to that level. Second initial threshold Or the third initial threshold (This serves as the base trigger threshold to be updated) Assume the normal cooling time is... Meanwhile, the system internally presets a high adoption threshold. The preset high ignore threshold is As a preferred parameter configuration method, the high adoption threshold is... The threshold can be set to the range of 0.80 to 0.90, indicating a high ignore threshold. It can be set to the range of 0.70 to 0.80.

[0083] The system is configured with a fixed threshold update step size parameter. Adjusting the step size with cooling time In this embodiment, the threshold update step size parameter The cooldown time adjustment step can be set from 0.02 to 0.05. The time can be set from 10 to 30 seconds. To prevent parameters from exceeding the logical range due to long-term accumulation or decay, the system synchronously presets the lower limit of the basic trigger threshold. and upper limit extreme value And the lower limit extreme value of the normal cooling time. and upper limit extreme value Based on the calculated statistical ratios, the system generates updated baseline trigger thresholds. And the updated regular cooldown time Its core update logic follows the piecewise function model: ; Corresponding new normal cooldown time The calculation logic is as follows: ; Based on the above judgment model, when a certain type of suggestion remains in a high adoption state for a long period of time ( When a certain type of prompt is highly ignored (e.g., when the step size parameter is reduced, the trigger threshold is lowered, and the normal cooldown time for such events is shortened to increase the frequency of information delivery), it indicates that the information is of high value to clinical decision-making. The system then lowers the trigger threshold by subtracting the step size parameter and simultaneously reduces the normal cooldown time for such events to increase the frequency of information delivery. Conversely, when a certain type of prompt is highly ignored (e.g., when the step size parameter is reduced, the trigger threshold is lowered), the trigger threshold is lowered. When an alarm is triggered (e.g., ) to indicate that it will interfere with normal medical treatment, the system raises the trigger threshold by increasing the step size parameter, simultaneously extending the regular cooldown time, or moving it to a background waiting queue. This is achieved by introducing a maximum value function. With minimum value function The system ensures that the parameters remain within a reasonable range regardless of the number of iterations. For the dynamic loading and network-wide distribution of updated threshold configurations, those skilled in the art can implement this using conventional microservice configuration center components, which are well-known technologies in the field and will not be elaborated upon here.

[0084] S630 implements high-risk lower limit protection and isolation defense rules. To prevent machine learning mechanisms based on statistical principles from undermining medical safety in extremely high-pressure scenarios, the system sets up a mandatory high-risk isolation defense line around the threshold learning engine. This rule explicitly stipulates that any medical event that ultimately reaches level R3 or triggers the high-risk lower limit direct access rule will be added to an exemption whitelist, restricting its participation in the aforementioned adjustments that would lead to an increase in the basic trigger threshold or an extension of the regular cooldown time.

[0085] For high-risk events falling within the isolation zone, the system only allows the optimization of terminal display channel parameters using behavioral feedback data. For example, the system can fine-tune the brightness contrast value of the flickering prompt at the edge of the field of view based on the light adaptability feedback from eye tracking; or adjust the frequency response curve of the bone conduction prompt based on the environmental noise sampling results. This high-risk lower limit protection mechanism blocks the possibility of risk underreporting from the system's underlying logic, avoiding the automatic shielding or excessive delay of truly high-risk events due to the long-term neglect preferences of some groups, thereby ensuring the long-term reliability and medical safety of this anti-interference control system.

[0086] Specific application examples: To further illustrate the collaborative operation of the various threshold parameters, control logic, and chart presentation effects in this application, the system operation mechanism of this application is explained in conjunction with specific application examples in the intensive care unit scenario and experimental verification data over a period of 5 weeks.

[0087] Clinical application examples and system operation process: Scenario: The attending physician is performing a central venous catheterization procedure on a patient. During this process, the bedside monitor detects that the patient's blood pressure rapidly drops from 115 / 75 mmHg to 80 / 45 mmHg. The system executes the following judgment and interruption control steps: Multimodal data acquisition and risk classification calculation (corresponding steps S10-S20): The AI ​​glasses' visual acquisition module and posture recognition module acquire multi-source heterogeneous data, and the event normalization module generates corresponding standardized medical events. The risk calculation module extracts the risk characteristic parameters of this standardized medical event: risk score. =0.88, urgency level =0.92, scene weight =0.85.

[0088] The system calculates the overall prompt score based on the formula. (The preset weighting factor is set to) , , ): ; Due to the comprehensive prompt score =0.886 and greater than the preset segmentation threshold (third initial threshold) (Set to 0.75), the basic warning level is mapped to R3. Simultaneously, this event hits the preset high-risk pass-through condition, and the system will lower the high-risk warning threshold. Set it to 3. Take the higher of the two values ​​to determine the final level. The value is determined to be 3.

[0089] Status recognition and unauthorized access trigger determination (corresponding to step S30): The posture recognition module determines the current aseptic operation state based on the collected data and sets the aseptic state parameters. =1. Based on the mapping function, the basic interruptible level. =1. Combined with the downgrade factor. Calculations are performed with a value of 0, and the level can be temporarily interrupted. .

[0090] The system executed an unauthorized access control trigger. This is an indicator of newly emerging critical sources in current standardized medical events. =0, actual risk score increment =0.10, the preset risk increment threshold =0.15, lower limit of high-risk warning =3. Unauthorized access trigger flag. The calculation process is as follows: ; Because it satisfies The condition for triggering the unauthorized access flag. =1.

[0091] The system calculates the final interruptible level based on this. : ; The system will eventually determine the interruptible level. =3 is added to the corresponding standardized medical event.

[0092] Instruction generation and execution (corresponding steps S40-S50): System parameters for determining sterility =1 and final warning level =3, prompting the queue module to generate a command to block the screen highlight pop-up, redirecting the display channel to play the prompt voice through the bone conduction unit and controlling the corner areas of the screen display unit to perform strobe. At the same time, the hospital server triggers an asynchronous evidence-based task to retrieve treatment suggestions and populate them back to the system interface.

[0093] Experimental verification of closed-loop feedback mechanism and explanation of graphic elements: To verify the effectiveness of step S60 in collecting interactive feedback data and updating the scoring threshold parameters, a comparative experiment was conducted for five consecutive weeks in a test environment. (Refer to...) Figure 3 The specific meanings and operational logic of each element in the diagram are as follows: Horizontal axis parameter description: The horizontal axis is defined as the "system operation and learning cycle (weeks)," with data ranging from week 1 to week 5. This axis represents the data accumulation and parameter update process of the feedback learning module within a fixed statistical time window.

[0094] Explanation of data on the left vertical axis and solid line: The left vertical axis is defined as "invalid alarm interruption rate (%)". The corresponding solid line (with a circular marker) in the legend shows that this indicator decreased sequentially from 68.5% in week 1 to 45.2%, 28.7%, 18.5%, and finally reached 12.3% in week 5.

[0095] Explanation of operational logic: The downward trend in this data corresponds to the result of the system's formula calculation. Within the statistical time window, the system calculates the ignore rate for specific types of events. (in To ignore the number of times, (Total number of times). When the ignore rate Greater than or equal to the preset high ignore threshold At that time, the system performs an update operation: ; ; Update the step size parameter by increasing the threshold. Adjusting the step size with cooling time The system outputs the updated base trigger threshold. Compared to the updated regular cooldown time Adjusting the parameters reduces the frequency of such high-ignore-rate events, thus resulting in a lower invalid alarm interruption rate in subsequent system cycles.

[0096] Explanation of the data on the right vertical axis and the dashed line: The right vertical axis is defined as "average response time (seconds) for high-risk events". The corresponding dashed line (marked with a square) in the legend shows that this indicator slightly decreased from 2.6s in week 1 to 2.4s and 2.2s, and then stabilized at 2.1s in weeks 4 and 5.

[0097] Operational logic explanation: The stable state of this data corresponds to the high-risk lower limit protection and isolation defense rules in the system. Based on these rules, the system ultimately alerts users to standardized medical events that reach level R3 or meet the high-risk direct access criteria (e.g., those described in the previous example). =3 events), restricted from participating in the aforementioned events. and The update calculation process is as follows. This protection mechanism ensures that the high-risk alert lower limit rule is not interfered with by behavioral feedback data, so that high-risk medical events maintain a consistent response priority output throughout the entire operation cycle. At the same time, the reduction in the invalid alarm interruption rate reduces the amount of concurrent information, making the execution of system resources and interactive actions more focused. This is reflected in the data as the average response time of high-risk events does not increase.

[0098] It should be understood that the terminology used herein is for the purpose of describing particular exemplary embodiments only and is not intended to be limiting. Unless the context clearly indicates otherwise, the singular forms “a,” “an,” and “described” as used herein may also include the plural forms. The terms “comprising,” “including,” “containing,” and “having” are inclusive and therefore indicate the presence of the stated features, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, elements, components, and / or combinations thereof. The method steps, processes, and operations described herein are not construed as requiring them to be performed in a particular order described or illustrated unless the order of performance is explicitly indicated. It should also be understood that additional or alternative steps may be used.

[0099] Although terms such as first, second, third, etc., may be used in this document to describe multiple elements, components, regions, layers, and / or segments, these elements, components, regions, layers, and / or segments should not be limited by these terms. These terms may be used only to distinguish one element, component, region, layer, or segment from another. Unless the context clearly indicates otherwise, terms such as "first," "second," and other numerical terms used herein do not imply order or sequence. Therefore, the first element, component, region, layer, or segment discussed below may be referred to as the second element, component, region, layer, or segment without departing from the teachings of the exemplary embodiments.

[0100] The above description is merely a specific embodiment of this application, enabling those skilled in the art to understand or implement this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features claimed herein.

Claims

1. A method for AI glasses medical risk classification prompts and interruption control, characterized in that, include: Acquire bedside multimodal data and instrument readings, align them according to timestamps, and encapsulate them into standardized medical events containing risk scores, urgency levels, scenario weights, and high-risk warning thresholds; The risk score, urgency level, and scenario weight are weighted together to obtain a comprehensive alert score. The comprehensive alert score is used to map a basic alert level. The larger value between the basic alert level and the lower limit of the high-risk alert is taken to determine the final alert level. Extract the field status parameters and the comprehensive confidence score of the status identification from the bedside multimodal data. The field status parameters include aseptic status parameters. Map the basic breakable level based on the field status parameters. Use the comprehensive confidence score of the status identification to correct the basic breakable level to obtain a temporary breakable level. Calculate the unauthorized triggering flag based on the actual risk score increment. Combine the temporary breakable level to establish the final breakable level. The standardized medical event is diverted and interruption control command is generated by using the final prompt level and the final interruptible level. The prompt output module is called to execute the output of the physical display channel, and the asynchronous evidence-based task is triggered to extract clinical guidelines and backfill evidence-based data.

2. The method for AI glasses medical risk classification prompting and interruption control according to claim 1, characterized in that, The step of determining the final alert level by taking the larger value between the basic alert level and the high-risk alert lower limit includes: Determine whether the standardized medical event meets the high-risk direct access condition or the continuous abnormal condition; When a high-risk direct access condition is met, the lower limit of the high-risk warning will be forcibly set to level R3. When a continuous abnormal condition is met, the lower limit of the high-risk warning will be forcibly set to level R2.

3. The method for AI glasses medical risk classification prompting and interruption control according to claim 2, characterized in that, The determination of the final warning level also includes: Map the basic prompt level to an integer constant in the memory structure; The larger of the two values ​​is used as the final warning level, by comparing the integer constant with the lower limit of the high-risk warning after being forcibly set.

4. The method for AI glasses medical risk classification prompting and interruption control according to claim 1, characterized in that, The step of using the state recognition comprehensive confidence level to correct the basic breakable level to obtain a temporary breakable level includes: Compare the overall confidence level of the state recognition with the state confidence threshold; When the overall confidence level of the state recognition is less than the state confidence threshold, a degradation factor is introduced. The numerical result is obtained by subtracting the degradation factor from the basic breakable level, and the maximum value between zero and the numerical result is determined as the temporary breakable level.

5. The method for AI glasses medical risk classification prompting and interruption control according to claim 1, characterized in that, The process of calculating the unauthorized trigger flag based on the actual risk score increment, and establishing the final interruptible level in conjunction with the temporary interruptible level, includes: Extract the new critical source identifiers of the standardized medical events; When the value of the newly emerging critical source identifier is equal to 1, or the actual risk score increment is greater than or equal to the risk increment threshold, or the value of the high-risk warning lower limit is equal to 3, the unauthorized trigger identifier is set to the value 1 corresponding to allowing unauthorized access. The product of the overreach trigger flag and the constant 3 is compared with the temporary interruptible level, and the larger of the two values ​​is determined as the final interruptible level.

6. The method for AI glasses medical risk classification prompting and interruption control according to claim 1, characterized in that, The process of performing event triage on the standardized medical event using the final alert level and the final interruptibility level includes: The standardized medical events are pushed into a waiting queue using the final alert level and the final interruptible level. Obtain the queue waiting time, source reliability coefficient, and maximum tolerable waiting time scale for the standardized medical events within the waiting queue; Calculate the ratio of the queue waiting time to the maximum tolerable waiting time scale, and introduce a constraint factor to limit the upper limit of the ratio to no more than one.

7. The AI ​​glasses medical risk classification prompt and interruption control method according to claim 6, characterized in that, The generation of interrupt control instructions includes: The final prompt level is quantified to obtain a scalar value for the final prompt level; The dynamic dequeue priority is obtained by weighting and summing the final prompt level scalar value, the urgency, the ratio, and the source reliability coefficient using different weighting coefficients. The interruption control command is then generated in descending order of the dynamic dequeue priority.

8. The method for AI glasses medical risk classification prompting and interruption control according to claim 1, characterized in that, The call prompt output module executes the output of the physical display channel, triggers an asynchronous evidence-based task to extract clinical guidelines and backfill evidence-based data, including: Extract the sterile state parameters. When the value of the sterile state parameter is equal to 1 or the value of the final interruptible level is less than or equal to 1, block the bright obscuring pop-up window and the conventional buzzer alarm channel in the interruption control command, and redirect the physical display channel to call the bone conduction unit to perform bone conduction voice broadcast and simultaneously call the screen display unit to perform micro-frequency flicker at the edge of the field of view. The recommended risk level associated with the clinical guidelines is analyzed, and the final prompt level is numerically processed to obtain a scalar value of the final prompt level. The difference between the recommended risk level and the scalar value of the final prompt level is compared. When the difference is greater than or equal to the conflict determination threshold, an abnormal conflict state is established. Under the abnormal conflict state, the output of the physical display channel that has been activated remains unchanged and an additional graphic message carrying a review label is generated.

9. An AI glasses medical risk grading prompt and interruption control system, characterized in that, The method for medical risk classification prompting and interruption control of AI glasses as described in any one of claims 1-8 includes an AI glasses terminal, an edge computing terminal, and an in-hospital server terminal connected by communication; The AI ​​glasses are used to acquire bedside multimodal data and send it to the edge computing terminal; The edge computing terminal is used to receive the bedside multimodal data and instrument readings, align them according to timestamps, and encapsulate them into standardized medical events containing risk scores, urgency levels, scenario weights, and high-risk warning limits; The risk score, urgency level, and scenario weight are weighted together to obtain a comprehensive alert score. The comprehensive alert score is used to map a basic alert level. The larger value between the basic alert level and the lower limit of the high-risk alert is taken to determine the final alert level. The edge computing terminal is also used to extract the field status parameters and the comprehensive confidence level of the status recognition of the bedside multimodal data. The field status parameters include sterile status parameters. The basic breakable level is mapped based on the field status parameters. The basic breakable level is corrected using the comprehensive confidence level of the status recognition to obtain a temporary breakable level. The unauthorized triggering flag is calculated based on the actual risk score increment. The final breakable level is determined by combining the temporary breakable level. The edge computing terminal is also used to perform event triage on the standardized medical event by utilizing the final prompt level and the final interruptible level, and generate an interruption control command to send to the AI ​​glasses terminal, triggering an asynchronous evidence-based task to the hospital server terminal. The AI ​​glasses are also used to receive the interruption control command and call the prompt output module to execute the output of the physical display channel. The in-hospital server is used to receive the asynchronous evidence-based tasks, extract clinical guidelines, and backfill the evidence-based data to the edge computing terminal.

10. The AI ​​glasses medical risk grading prompt and interruption control system according to claim 9, characterized in that, The edge computing terminal is specifically used to extract the aseptic state parameters; When the sterile state parameter value is equal to 1 or the final interruptible level value is less than or equal to 1, the high-brightness blocking pop-up and the conventional buzzer alarm channel are blocked in the interruption control command, and the physical display channel is redirected to call the bone conduction unit to perform bone conduction voice broadcast and simultaneously call the screen display unit to perform micro-frequency flicker at the edge of the field of view. Specifically, the AI ​​glasses are used to execute corresponding output actions through the built-in bone conduction unit and screen display unit based on the received interruption control command after redirection.

Images