An endoscopy regurgitation aspiration risk management interactive method and system

Abstract

The application provides a gastroscopy regurgitation and aspiration risk management interaction method and system, relates to the technical field of intelligent medical treatment, and the method is convenient for doctors to efficiently operate by uniformly integrating various interface entrances through a main interface. The method guides the entry of individualized information of a patient before an operation, intelligently recommends default early warning rules based on the information, and allows doctors to flexibly adjust according to clinical judgment, thereby improving the accuracy of risk configuration. During the examination process, a monitoring instrument panel presents multi-source monitoring data in real time, dynamically generates a regurgitation and aspiration risk level by combining preoperative information with a risk assessment model, and once a preset early warning rule is triggered, a risk intervention suggestion window is immediately popped up on the interface, specific executable clinical measures are pushed, and doctor feedback is collected, so that man-machine collaborative decision-making is realized. After the examination is completed, the risk log sub-interface supports the backtracking of risk assessment records, intervention interaction information and doctor feedback in the examination process, so as to support the continuous optimization of the risk assessment model.

Description

Technical Field

[0001] This application relates to the field of smart healthcare technology, and in particular to an interactive method and system for managing the risk of reflux aspiration during gastroscopy. Background Technology

[0002] Gastroscopy under sedation or anesthesia is prone to reflux and aspiration, which can lead to serious complications and even death, especially in high-risk groups such as obese individuals and those with diabetes. Current clinical practice relies primarily on static indicators such as fasting time to assess risk, but research shows that fasting duration does not accurately reflect gastric emptying status, and individual differences are significant. Existing methods, such as ultrasound measurement of antral area or electronic medical record screening, have some application, but the data is scattered across multiple isolated systems, lacking integration and dynamic updates. Risk assessment relies heavily on physician experience, lacking standardized and quantitative models, and is often a one-time preoperative assessment, unable to adjust in real time according to dynamic factors such as intraoperative gas injection and changes in patient position. Even with the introduction of artificial intelligence, it often only outputs a general risk level, lacking interpretability and specific intervention suggestions, with weak human-computer interaction and difficulty in closed-loop optimization. Overall, existing technologies suffer from shortcomings such as single assessment dimensions, data silos, lack of dynamic response, and insufficient clinical applicability. Summary of the Invention

[0003] This application provides an interactive method and system for managing the risk of reflux aspiration during gastroscopy, in order to solve one or more technical problems existing in the prior art, and at least provide a beneficial option or create conditions that enable closed-loop interactive management of multi-source data fusion, dynamic risk assessment and interpretable intervention recommendations.

[0004] On the one hand, this application provides an interactive method for managing the risk of endoscopic reflux aspiration, including the following steps: The main interface for managing the risk of gastroscopy reflux aspiration includes preoperative information entry controls, risk warning configuration controls, a monitoring dashboard area, and risk log viewing controls. In response to a trigger command on the preoperative information entry control, a preoperative information entry sub-interface is displayed to enter relevant preoperative information of the patient; In response to the trigger command of the risk warning configuration control, a risk warning configuration sub-interface is displayed, providing intelligently recommended default warning rules, and supporting the adjustment of the default warning rules based on the preoperative information to complete the configuration of warning rules for the risk of reflux aspiration. During the gastroscopy, the monitoring data is displayed in real time through the monitoring dashboard area. Based on the preoperative information and the monitoring data, a real-time risk assessment result for reflux aspiration is generated using a risk assessment model. When the risk assessment result and / or the monitoring data meet the warning rules, a risk intervention suggestion window automatically pops up on the main interface, pushes clinical intervention suggestions, and receives feedback from doctors. In response to the trigger command of the risk log viewing control, the risk log sub-interface is displayed, which supports retrospective examination of risk assessment records, intervention interaction information and doctor feedback, so as to support the continuous optimization of the risk assessment model.

[0005] Furthermore, the preoperative information entry sub-interface includes a patient information entry control, a medical history and comorbidity configuration control, a fasting time setting control, a gastric contents status entry control, and a preoperative information saving control; On the preoperative information entry sub-interface, enter the patient's relevant preoperative information, including the following steps: In response to a trigger command on the patient information entry control, a patient information input window is displayed, and the patient information entered includes the patient's age, gender, weight, height, and corresponding body mass index; In response to a trigger command to the medical history and comorbidity configuration control, a comorbidity selection window is displayed, and medical history items related to the risk of reflux and aspiration are selected, including at least one of diabetes, gastroparesis, hiatal hernia, or a history of aspiration. In response to the trigger command of the fasting time setting control, the system inputs the time point of the patient's last meal or drink, and automatically calculates the actual fasting duration; In response to a trigger command to the gastric contents status recording control, gastric contents status information is recorded, including at least one of the following: gastric antrum cross-sectional area CSA value, Perlas grade, or "not evaluated" label; In response to a trigger command to the preoperative information saving control, the recorded patient information, medical history and comorbidities, fasting time and gastric contents status information are saved, and a dataset of patient-related preoperative information is generated.

[0006] Furthermore, the risk warning configuration sub-interface includes an intelligent recommendation control, a threshold setting control, an alarm mode configuration control, and a rule saving control; In the risk warning configuration sub-interface, configuring the warning rules includes the following steps: In response to the trigger command of the intelligent recommendation control, a default warning rule matching the patient's risk characteristics is automatically loaded based on the preoperative information; In response to a trigger command on the threshold setting control, a threshold setting window is displayed, allowing modification of one or more conditions in the default warning rules. The conditions include a risk probability threshold, an intraoperative operation parameter threshold, or a physiological parameter change threshold. The intraoperative operation parameter threshold includes a carbon dioxide injection rate threshold or a cumulative fluid injection volume threshold, and the physiological parameter change threshold includes a threshold for the decrease in blood oxygen saturation SpO2 per unit time. In response to a trigger command to the alarm mode configuration control, a prompting mode is configured when a risk is triggered, wherein the prompting mode includes at least one of pop-up reminder, sound alarm, or interface highlighting; In response to the trigger command of the rule saving control, the finally confirmed early warning rule and the corresponding alarm method are saved.

[0007] Furthermore, the step of generating a real-time risk assessment result for reflux aspiration based on the preoperative information and the monitoring data using a risk assessment model includes the following steps: In response to the gastroscopy start command, the risk assessment model is initialized and the preoperative information is loaded as static input features; During the gastroscopy, intraoperative operation data from the endoscope host and physiological parameters from the monitoring equipment are continuously collected at preset time intervals as dynamic input features; The static input features and the dynamic input features are adaptively fused using a self-attention mechanism to generate a weighted fused temporal feature representation; wherein, the self-attention mechanism dynamically allocates fusion weights based on the correlation between each feature and the current risk of backflow aspiration. The weighted and fused temporal feature representation is input into the risk assessment model, and the probability of backflow and mis-absorption risk at the current moment is output. Based on the probability of backflow and accidental aspiration, the corresponding risk level is determined and updated and displayed in real time in the monitoring dashboard area.

[0008] Furthermore, the risk intervention suggestion window includes a structured suggestion display area, an interpretability analysis area, a suggestion adoption control, and a suggestion rejection control; During a gastroscopy, when the risk assessment results and / or the monitoring data meet the warning rules, a risk intervention suggestion window automatically pops up, and clinical intervention suggestions are pushed and doctor feedback is received, including the following steps: In response to the determination that the risk triggering conditions are met, the risk intervention suggestion window is activated, and the current risk level and the corresponding structured clinical intervention suggestions are displayed on the main interface. In the structured suggestion display area, a list of intervention measures matching the current risk scenario is displayed. The intervention measures include at least one of suspending carbon dioxide injection, reducing the amount of fluid injected, adjusting the patient's position, or strengthening airway monitoring. In the interpretability analysis area, the contribution of each input feature to the current risk assessment result is visualized to generate an interpretability analysis report; In response to the trigger command of the suggestion adoption control, the doctor's confirmation of the implementation of the intervention suggestion is recorded, and the operation is associated with the current risk event log; In response to the trigger command of the suggestion rejection control, a remarks input window pops up, receives the rejection reason entered by the doctor, and saves the rejection operation and reason to the feedback record for subsequent iterative optimization of the risk assessment model.

[0009] Furthermore, the risk log sub-interface includes a risk event timeline control, an intervention record details control, a doctor feedback viewing control, and a model optimization status indicator control; The risk log sub-interface supports retrospective review of risk assessment records, intervention interaction information, and doctor feedback during the examination process, including the following steps: In response to the trigger command of the risk event timeline control, risk event nodes are displayed in chronological order, with each node associated with the real-time risk probability, risk level, and triggering warning rules at the corresponding moment; In response to the trigger command of the intervention record details control, the structured clinical intervention suggestion content and push time corresponding to the selected risk event node are displayed; In response to a trigger command to view the doctor's feedback control, the doctor's operation record for the intervention suggestion is displayed, including acceptance confirmation or rejection notes and the reasons for input; In response to the trigger command of the model optimization status indicator control, the progress status or version update information of the iterative optimization of the risk assessment model based on the feedback records accumulated in this inspection is displayed.

[0010] Furthermore, the risk assessment model adopts an edge-cloud collaborative architecture, including a global large model deployed in the cloud and a lightweight model deployed on the edge computing device at the gastroscopy examination site; wherein, the global large model is trained based on multi-center historical case data, and the lightweight model is generated from the global large model through model compression or knowledge distillation, and deployed on the edge computing device to support local real-time inference.

[0011] Furthermore, the method further includes the following steps: After the gastroscopy is completed on the edge computing device, the structured feedback data from the examination is encrypted and uploaded to the cloud. In the cloud, the global large model is periodically retrained based on newly collected feedback data; The updated global model is compressed or distilled again into a new lightweight model and distributed to various edge computing devices to achieve the continuous collaborative evolution of the risk assessment model.

[0012] Furthermore, the monitoring dashboard area is also used to display the risk assessment results of postoperative reflux aspiration predicted by the risk assessment model, specifically including: In response to the end command of the gastroscopy, based on the intraoperative operation data accumulated during the examination, the trend of physiological parameter changes and the final gastric state, the risk assessment model is used to infer the probability of delayed gastric emptying and the recovery time of airway protection ability during the postoperative recovery period. Based on the gastric emptying delay probability and the airway protection recovery time, a postoperative reflux aspiration risk score is generated. The postoperative reflux and aspiration risk score is displayed as an independent postoperative risk card in the monitoring dashboard area, and high-risk time periods are marked. When the postoperative reflux and aspiration risk score exceeds a preset threshold, postoperative care suggestions are pushed to the main interface. The postoperative care suggestions include at least one of maintaining a lateral decubitus position, delaying drinking and eating, strengthening SpO2 monitoring, or extending the PACU observation time.

[0013] On the other hand, this application provides an interactive system for managing the risk of endoscopic reflux aspiration, including the following modules: The main interface module is configured to display the main interface for managing the risk of gastroscopy reflux aspiration. The main interface includes preoperative information entry controls, risk warning configuration controls, a monitoring dashboard area, and risk log viewing controls. The preoperative information entry module is configured to: in response to a trigger command on the preoperative information entry control, display a preoperative information entry sub-interface and enter relevant preoperative information of the patient; The risk warning configuration module is configured to: in response to the trigger command of the risk warning configuration control, display the risk warning configuration sub-interface, provide intelligently recommended default warning rules, and support the adjustment of the default warning rules based on the preoperative information to complete the configuration of the warning rules for the risk of reflux aspiration. The monitoring dashboard module is configured to: display monitoring data in real time through the monitoring dashboard area during the gastroscopy examination, and generate a real-time risk assessment result of reflux aspiration based on the preoperative information and the monitoring data using a risk assessment model; when the risk assessment result and / or the monitoring data meet the warning rules, a risk intervention suggestion window will automatically pop up on the main interface, push clinical intervention suggestions and receive doctor feedback; The risk log viewing module is configured to display a risk log sub-interface in response to a trigger command on the risk log viewing control, supporting the retrospective review of risk assessment records, intervention interaction information, and doctor feedback during the examination process, in order to support the continuous optimization of the risk assessment model.

[0014] The beneficial effects of this application are as follows: This application provides an interactive method for managing the risk of reflux aspiration during gastroscopy. This method integrates preoperative information input controls, risk warning configuration controls, a monitoring dashboard area, and a risk log viewing control through a unified main interface, facilitating efficient operation by doctors. The method guides the input of individualized patient information before the procedure and intelligently recommends default warning rules based on this information. It also allows doctors to flexibly adjust the settings according to clinical judgment, improving the accuracy of risk configuration. During the examination, the monitoring dashboard displays multi-source monitoring data in real time. Combined with preoperative information, the risk assessment model dynamically generates a reflux aspiration risk level. Once a preset warning rule is triggered, a risk intervention suggestion window immediately pops up on the interface, pushing specific executable clinical measures and collecting doctor feedback, achieving human-machine collaborative decision-making. After the examination, the risk log sub-interface allows for reviewing risk assessment records, intervention interaction information, and doctor feedback during the examination process to support continuous optimization of the risk assessment model. This application also provides a corresponding system; the beneficial effects of the system are similar to the method and will not be elaborated further here.

[0015] Other features and advantages of this application will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the application. The objectives and other advantages of this application may be realized and obtained by means of the structures particularly pointed out in the description, claims and drawings. Attached Figure Description

[0016] The accompanying drawings are provided to further understand the technical solutions of the present invention and constitute a part of the specification. They are used together with the embodiments of the present invention to explain the technical solutions of the present invention, and do not constitute a limitation on the technical solutions of the present invention.

[0017] Figure 1 This is a flowchart of the interactive method for managing the risk of reflux aspiration during gastroscopy provided in this application; Figure 2 This is a schematic diagram of the main interface for the gastroscopy reflux aspiration risk management interaction provided in this application; Figure 3 This is a schematic diagram of the preoperative information entry sub-interface provided in this application; Figure 4 This is a schematic diagram of the risk warning configuration sub-interface provided in this application; Figure 5 This is a schematic diagram of the risk intervention suggestion window provided in this application; Figure 6 This is a schematic diagram of the risk log sub-interface provided in this application; Figure 7 This is a structural diagram of the gastroscopy reflux aspiration risk management interactive system provided in this application. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0019] The present application will be further described below with reference to the accompanying drawings and specific embodiments. The described embodiments should not be considered as limitations on the present application, and all other embodiments obtained by those skilled in the art without inventive effort are within the scope of protection of the present application.

[0020] In the following description, references are made to “some embodiments,” which describe a subset of all possible embodiments. However, it is understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0021] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0022] Gastroscopy, a crucial tool for diagnosing and treating upper gastrointestinal diseases, is widely used globally, with tens of millions performed annually. In clinical practice, to improve patient comfort and reduce operative interference, most gastroscopy procedures are performed under sedation or general anesthesia. However, this significantly suppresses the patient's swallowing reflex, cough reflex, and lower esophageal sphincter tone, thereby greatly increasing the risk of gastric contents reflux into the esophagus and aspiration into the airway. Once reflux aspiration occurs, it can lead to chemical pneumonia, acute respiratory distress syndrome, and in severe cases, even death. Although the reported clinically significant incidence is 0.01% to 0.1%, the risk is significantly increased in high-risk groups such as those with obesity, diabetes, gastroparesis, pregnancy, neurological disorders, or those undergoing emergency surgery. Furthermore, the actual incidence may be underestimated due to missed diagnoses or lack of documentation.

[0023] Currently, the primary basis for assessing the risk of reflux and aspiration during gastroscopy in clinical practice is the perioperative fasting guidelines published by anesthesiology societies in various countries, which typically recommend fasting for six hours on solid food and two hours on clear liquids preoperatively. This strategy has long been standard practice, but numerous studies have confirmed that fasting time does not reliably reflect an individual's true gastric emptying status. In some patients, even after strictly fasting for more than eight hours, significant gastric residue can still be detected on ultrasound; while in others, the stomach is essentially empty after a shorter fast. This indicates that relying solely on fasting duration for risk assessment has significant limitations and fails to reflect individual differences in gastrointestinal motility, metabolic status, and underlying diseases.

[0024] Besides fasting time, factors affecting the risk of reflux aspiration include changes in intragastric pressure, the amount of carbon dioxide or air injected during endoscopy, the amount of fluid injected, patient position, depth of anesthesia, degree of swallowing reflex inhibition, and the presence of increased intra-abdominal pressure (such as ascites or intestinal obstruction). These factors are mostly dynamic variables that evolve in real time during the examination, but traditional assessment methods lack the ability to quantitatively collect and integrate these key parameters.

[0025] In recent years, some studies have attempted to introduce technological means to improve the objectivity of risk assessment. For example, some teams have used bedside abdominal ultrasound to measure the cross-sectional area of ​​the gastric antrum and combined it with parameters such as age and body mass index to establish regression models to estimate gastric contents; other medical institutions have embedded preoperative risk screening forms into electronic medical record systems, with nurses or anesthesiologists manually selecting high-risk factors; and high-end endoscopy systems have also begun to record the total amount of gas and water used during the procedure. However, these technologies remain isolated, static, and not in a closed-loop state. Ultrasound measurements rely on operator experience and are difficult to monitor continuously; screening forms in electronic medical records are mostly qualitative judgments, lacking quantitative thresholds; and endoscopy system data is not linked to the patient's physiological state or risk model. More importantly, the above information is scattered across multiple independent systems such as ultrasound equipment, electronic medical record systems, endoscopy systems, and monitors, forming data silos that require doctors to manually summarize, which is not only inefficient but also prone to missing key risk factors.

[0026] Therefore, existing technologies have six core shortcomings in managing the risk of reflux aspiration during gastroscopy. First, the assessment dimensions are too singular, relying excessively on static indicators such as fasting time, neglecting individual differences and dynamic changes in intragastric pressure and gas injection volume during the procedure. Second, data sources are scattered, lacking a unified platform to integrate multi-source heterogeneous data, resulting in incomplete risk assessment information. Third, risk judgment is highly dependent on physicians' personal experience, lacking standardized, quantifiable intelligent assessment models trained on large samples, leading to significant differences in judgments among different doctors and poor consistency. Fourth, existing methods are mostly one-time preoperative assessments, unable to dynamically reassess based on real-time intraoperative signals, resulting in delayed interventions and difficulty in timely warnings at the initial stage of risk escalation. Fifth, even when some systems incorporate artificial intelligence models, they often only output high, medium, and low risk labels without explaining the basis for the judgment or providing specific, actionable clinical intervention suggestions, limiting their practicality. Sixth, human-computer interaction is weak; doctors cannot confirm, correct, or supplement system suggestions, and the system cannot learn and optimize from clinical feedback, resulting in a disconnect between the model and real-world clinical scenarios, hindering continuous evolution.

[0027] To address the aforementioned issues, this application proposes an interactive method and system for managing the risk of reflux aspiration during gastroscopy, constructing an intelligent closed-loop management platform covering the entire process from pre-operative to post-operative stages. This method integrates four main functional modules through the main interface: pre-operative information entry, risk warning rule configuration, real-time monitoring dashboard, and risk log viewing, achieving unified aggregation and collaborative operation of multi-source clinical data. Pre-operatively, the system intelligently recommends default warning rules based on individual patient characteristics and supports dynamic adjustments by physicians based on clinical experience, improving the personalization and accuracy of risk configuration. During the gastroscopy, the system collects dynamic information such as endoscopic operation parameters and physiological monitoring data in real time. Combined with pre-operative data, it drives a risk assessment model to continuously output the real-time risk level of reflux aspiration. When a preset rule is triggered, a suggestion window containing specific intervention measures automatically pops up, while simultaneously receiving physician feedback to achieve human-machine collaborative decision-making. After the examination, all risk assessment records, intervention interactions, and physician feedback are structured and stored in the risk log, providing high-quality closed-loop data for continuous model learning and optimization, thereby continuously improving the system's accuracy, interpretability, and clinical adaptability.

[0028] First, the interactive method for managing the risk of reflux aspiration during gastroscopy provided in the embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0029] Reference Figure 1 The implementation process of the gastroscopy reflux aspiration risk management interactive method provided in this application embodiment includes, but is not limited to, the following steps.

[0030] Step S110: Display the main interface 100 of the gastroscopy reflux aspiration risk management interaction.

[0031] Among them, reference Figure 2 The main interface 100 includes a preoperative information entry control 101, a risk warning configuration control 102, a monitoring dashboard area 103, and a risk log viewing control 104.

[0032] In step S110, a unified operation entry point and visual hub are provided for the entire gastroscopy reflux aspiration risk management interactive method. By displaying the main interface 100, the system centrally presents core functional modules in the form of controls, including preoperative information entry controls 101, risk warning configuration controls 102, monitoring dashboard area 103, and risk log viewing controls 104. This integrated layout allows doctors or operators to quickly access the functions required at different stages on a single interface, avoiding switching between multiple independent systems, thereby improving operational efficiency and usability. The main interface 100, as the central control platform for human-computer interaction, not only organizes the trigger paths for subsequent steps but also establishes a structured operational framework for end-to-end risk management, serving as the fundamental carrier for achieving closed-loop management.

[0033] In step S120, in response to the trigger command of the preoperative information entry control 101, the preoperative information entry sub-interface 200 is displayed to enter the patient's relevant preoperative information.

[0034] In step S120, the data foundation for individualized risk assessment is established. When the user triggers the preoperative information entry control 101, the system displays the preoperative information entry sub-interface 200, used to collect preoperative clinical data closely related to the patient's risk of reflux and aspiration. This information may include key variables such as age, body mass index, underlying diseases such as diabetes or neurological disorders, fasting time, and gastric emptying-related symptoms. Through structured entry, the system ensures that the risk assessment model has accurate and complete input parameters in subsequent operations, thereby improving the relevance and reliability of predictions. This step transforms information originally scattered in medical records or verbal interviews into calculable and traceable standardized data, which is a prerequisite for achieving accurate risk identification.

[0035] In step S130, in response to the trigger command of the risk warning configuration control 102, the risk warning configuration sub-interface 300 is displayed, providing intelligently recommended default warning rules and supporting the adjustment of the default warning rules based on preoperative information to complete the configuration of warning rules for the risk of reflux aspiration.

[0036] In step S130, the personalized setting and clinical adaptation of risk warning thresholds are implemented. After the user triggers the risk warning configuration control 102, the system displays the risk warning configuration sub-interface 300 and intelligently recommends a set of default warning rules based on the entered preoperative information. This default rule is generated by a pre-trained risk model and reflects the intervention threshold under typical high-risk characteristics. At the same time, the system allows doctors to adjust the default rule according to the specific patient's condition or their own clinical judgment, such as modifying the intragastric pressure warning value, the upper limit of gas injection volume, or the risk score trigger point. This mechanism retains the objective recommendation ability of artificial intelligence while respecting clinical professional judgment, making the warning rules both scientific and flexible, and effectively avoiding false alarms or omissions caused by "one-size-fits-all" assessments.

[0037] In step S140, during the gastroscopy, monitoring data is displayed in real time on the monitoring dashboard area 103. Based on preoperative information and monitoring data, a real-time risk assessment result for reflux aspiration is generated using a risk assessment model. When the risk assessment result and / or monitoring data meet the warning rules, a risk intervention suggestion window 400 automatically pops up on the main interface 100, pushing clinical intervention suggestions and receiving feedback from doctors.

[0038] In step S140, dynamic risk perception and real-time intervention guidance are achieved during the gastroscopy procedure. During the operation, the monitoring dashboard 103 continuously displays monitoring data from the endoscopic equipment, monitor, or other sensing systems, such as CO2 injection volume, water injection volume, changes in body position, and intragastric pressure trends. The system inputs this real-time data along with preoperative information into the risk assessment model, continuously generating updated reflux aspiration risk assessment results. Once this result or the original monitoring data reaches or exceeds the threshold set by the previously configured warning rules, the main interface 100 immediately and automatically pops up a risk intervention suggestion window 400, pushing specific clinical response measures to the doctor, such as adjusting body position, pausing gas injection, speeding up the operation, or preparing the suction device, and simultaneously receiving the doctor's confirmation, disregard, or supplementary feedback on the suggestions. This step achieves a key shift from passive observation to proactive warning, and from static assessment to dynamic response.

[0039] In step S150, in response to the trigger command of the risk log viewing control 104, the risk log sub-interface 500 is displayed, which supports the retrospective examination process of risk assessment records, intervention interaction information and doctor feedback, so as to support the continuous optimization of the risk assessment model.

[0040] In step S150, a feedback loop for model optimization and knowledge accumulation is constructed. When the user clicks the risk log viewing control 104, the system displays the risk log sub-interface 500, which fully presents all key records related to risk during this gastroscopy, including risk assessment results at each time point, triggered warning events, intervention suggestions pushed by the system, and the doctor's actual feedback behavior. These structured logs not only support post-event review and quality assessment, but more importantly, they provide real-world labeled data for the iterative training of the risk assessment model. By continuously accumulating multiple interaction logs, the system can identify model biases, verify the effectiveness of interventions, and drive the algorithm to self-optimize, thereby continuously improving the accuracy and clinical fit of future risk predictions.

[0041] In some embodiments of this application, reference is made to Figure 3 The preoperative information entry sub-interface 200 includes a patient information entry control 201, a medical history and comorbidity configuration control 202, a fasting time setting control 203, a gastric contents status entry control 204, and a preoperative information saving control 205. In step S120, the patient's relevant preoperative information is entered in the preoperative information entry sub-interface 200, including the following steps.

[0042] In step S210, in response to the trigger command of the patient information input control 201, the patient information input window is displayed, and the patient information is entered, including the patient's age, gender, weight, height and corresponding body mass index.

[0043] In step S210, basic demographic and physiological parameters of the patient are collected to provide a foundational quantitative basis for subsequent reflux and aspiration risk assessment. When the user triggers the patient information entry control 201, the system displays a patient information input window, guiding the operator to enter the patient's age, gender, weight, height, and body mass index (BMI) automatically calculated by the system. These data are important variables affecting gastric emptying capacity, intra-abdominal pressure levels, and anesthesia tolerance. BMI, as an objective indicator of obesity, has been confirmed by multiple studies to be significantly associated with the risk of reflux and aspiration. By collecting this information in a structured manner, the system ensures that the risk assessment model can perform initial stratification based on standardized, computable individual characteristics, avoiding reliance on subjective judgment or omission of key baseline data.

[0044] In step S220, in response to the trigger command of the medical history and comorbidity configuration control 202, a comorbidity selection window is displayed, and medical history items related to the risk of reflux and aspiration are selected. The medical history items include at least one of diabetes, gastroparesis, hiatal hernia, or a history of aspiration.

[0045] In step S220, high-risk underlying diseases or past medical history with a clear pathophysiological association to reflux aspiration are identified and recorded. After the user triggers the medical history and comorbidity configuration control 202, the system displays a comorbidity selection window, allowing the operator to select at least one relevant medical history item from a preset list, including diabetes, gastroparesis, hiatal hernia, or a history of aspiration. These diseases significantly increase the likelihood of aspiration by delaying gastric emptying, weakening the lower esophageal sphincter, altering anatomical structures, or indicating a previous high-risk condition. This step transforms clinical experience into risk factor labels that can be recognized by the model, enabling the system to automatically assign higher weights to patients with specific pathological backgrounds, thereby improving the sensitivity and specificity of risk identification.

[0046] In step S230, in response to the trigger command of the fasting time setting control 203, the time point of the patient's last food or water intake is entered, and the actual fasting time is automatically calculated.

[0047] In step S230, the system accurately obtains the patient's actual fasting duration—a traditional but crucial risk reference indicator—and transforms it into structured time data. When the user triggers the fasting time setting control 203, the system allows input of the specific time point of the patient's last solid food intake or clear liquid consumption, and automatically calculates the actual fasting duration based on the current system time. Although fasting duration itself is not an absolutely reliable indicator of gastric emptying, it remains a widely used preliminary screening criterion in clinical practice when other dynamic monitoring methods are lacking. Through automated calculation, this step avoids errors from manual estimation and provides accurate time dimension input for subsequent joint analysis with other risk factors.

[0048] In step S240, in response to a trigger command to the gastric contents status recording control 204, gastric contents status information is recorded, including at least one of the following: gastric antrum cross-sectional area CSA value, Perlas grade, or "not evaluated" label.

[0049] In step S240, objective or semi-quantitative assessment results of gastric contents are introduced to compensate for the shortcomings of relying solely on fasting time. In response to a trigger command on the gastric contents status input control 204, the system allows the input of gastric contents status information, including the antral cross-sectional area (CSA) value obtained through ultrasound measurement, the Perlas classification based on ultrasound images, or a flag indicating no assessment was performed. Antral cross-sectional area and Perlas classification are currently the mainstream methods for bedside ultrasound assessment of gastric contents, with good reproducibility and clinical validation. This step allows the system to directly utilize observational data reflecting the actual residual state within the stomach, significantly improving the real-time nature and accuracy of risk assessment, especially suitable for high-risk or emergency patients.

[0050] In step S250, in response to the trigger command of the preoperative information saving control 205, the entered patient information, medical history and comorbidities, fasting time and gastric contents status information are saved, and a dataset of preoperative information related to the patient is generated.

[0051] In step S250, the structured integration and persistent storage of preoperative information are completed, providing a complete input dataset for subsequent risk modeling and process execution. When the user triggers the preoperative information save control 205, the system summarizes and saves the previously entered patient basic information, medical history and comorbidity options, actual fasting duration, and gastric contents status information, generating a complete patient-related preoperative information dataset. This dataset not only serves as a direct input source for risk warning rule configuration and real-time evaluation model operation during this examination, but also provides a standardized data foundation for subsequent log recording, model training, and quality backtesting. This step ensures that all previously collected information is effectively solidified, preventing data loss or inconsistency, and is a key link in achieving full-process traceable risk management.

[0052] In some embodiments of this application, reference is made to Figure 4 The risk warning configuration sub-interface 300 includes an intelligent recommendation control 301, a threshold setting control 302, an alarm mode configuration control 303, and a rule saving control 304. In step S130, the warning rules are configured in the risk warning configuration sub-interface 300, including the following steps.

[0053] In step S310, in response to the trigger command of the intelligent recommendation control 301, a default warning rule matching the patient's risk characteristics is automatically loaded based on preoperative information.

[0054] In step S310, the initial intelligent adaptation of the warning rules is achieved, improving configuration efficiency and clinical relevance. When the user triggers the intelligent recommendation control 301, the system automatically loads a set of default warning rules that match the current patient's risk characteristics based on the entered preoperative information. This default rule is generated by a pre-trained risk assessment model in the background, comprehensively considering factors such as the patient's age, body mass index, comorbidities, fasting duration, and gastric contents status, reflecting the key threshold combination most likely to trigger aspiration risk in this patient group during gastroscopy. Through this step, the system avoids doctors setting complex parameters from scratch, lowering the usage threshold, while ensuring that the initial rules are medically reasonable and individually targeted, laying a scientific foundation for subsequent personalized adjustments.

[0055] In step S320, in response to a trigger command on the threshold setting control 302, a threshold setting window is displayed, allowing modification of one or more conditions in the default warning rules, including risk probability threshold, intraoperative operation parameter threshold, or physiological parameter change threshold.

[0056] Among them, the intraoperative operation parameter thresholds include the CO2 injection rate threshold or the cumulative water injection volume threshold, and the physiological parameter change thresholds include the threshold for the decrease in blood oxygen saturation SpO2 per unit time.

[0057] It should be noted that SpO2 refers to the percentage of oxygen-bound hemoglobin in the blood, reflecting the blood's oxygen-carrying capacity and tissue oxygenation status. Clinically, SpO2 is an important vital sign indicator, typically measured non-invasively and in real-time using a fingertip pulse oximeter.

[0058] In normal adults at rest, SpO2 is generally maintained between 95% and 100%. When SpO2 falls below 90%, it usually indicates hypoxemia, which may be caused by respiratory diseases, circulatory disorders, airway obstruction, or acute events such as aspiration. During gastroscopy, if gastric contents are aspirated into the airway, it can lead to ventilation / perfusion mismatch and even alveolar collapse, causing a rapid drop in SpO2. Therefore, the magnitude of the rapid decrease in SpO2 per unit time is often used as an important physiological parameter for identifying early aspiration or respiratory depression.

[0059] In step S320, clinicians are granted flexible adjustment permissions for the core parameters of the warning rules to balance model recommendations with individualized treatment needs. After the user triggers the threshold setting control 302, the system displays a threshold setting window, allowing modification of one or more conditions in the default warning rules, including risk probability thresholds, intraoperative operation parameter thresholds, or physiological parameter change thresholds. The intraoperative operation parameter thresholds specifically include the carbon dioxide injection rate threshold and the cumulative fluid injection volume threshold, both directly related to increased intragastric pressure and the degree of gastric distension. The physiological parameter change thresholds include the threshold for the decrease in blood oxygen saturation (SpO2) per unit time, used to detect early aspiration-induced oxygenation deterioration. This step allows doctors to fine-tune the sensitivity based on the patient's real-time status, the operation plan, or their own experience, preventing excessive or missed alarms and enhancing the system's usability and reliability.

[0060] Step S330: In response to the trigger command of the alarm mode configuration control 303, configure the prompt mode when the risk is triggered. The prompt mode includes at least one of pop-up reminder, sound alarm or interface highlight.

[0061] In step S330, on-demand configuration of multimodal risk alert methods is supported to ensure that warning information can be perceived and responded to in a timely manner. When the user triggers the alarm method configuration control 303, the system allows selection of the alert method when a risk is triggered. This method includes at least one of pop-up reminders, sound alarms, or interface highlighting. Pop-up reminders can forcibly interrupt the current operation to attract attention, sound alarms are suitable for scenarios where the operator's eyes are not focused on the screen, and interface highlighting provides non-invasive visual cues. Different alert methods are suitable for different clinical environments and operational stages. This step enhances the adaptability of human-computer interaction by giving users control over the form of alarms, ensuring that warning information is not ignored at critical moments, thereby guaranteeing the timeliness of intervention.

[0062] Step S340: In response to the trigger command of the rule saving control 304, save the finally confirmed early warning rule and the corresponding alarm method.

[0063] In step S340, the final early warning strategy, confirmed by the doctor, is solidified, providing a clear basis for subsequent real-time monitoring and risk response. After the user triggers the rule save control 304, the system completely saves the early warning rules currently adjusted in the sub-interface and their corresponding alarm methods. The saved content includes all modified threshold conditions, selected risk trigger logic, and configured prompt formats, forming a complete and executable personalized early warning scheme. This step marks the formal completion of the early warning configuration process, ensuring that the system can strictly monitor and judge according to this rule during gastroscopy. It is a key link connecting preoperative preparation and intraoperative dynamic response, and also provides accurate rule context for subsequent log recording and model optimization.

[0064] In some embodiments of this application, step S140 involves generating a real-time risk assessment result for reflux aspiration based on preoperative information and monitoring data using a risk assessment model, including the following steps.

[0065] Step S410: In response to the gastroscopy start command, initialize the risk assessment model and load preoperative information as static input features.

[0066] In step S410, an initial state is established for the operation of the risk assessment model, and individualized static background information about the patient is injected. Upon receiving the gastroscopy initiation command, the system immediately initializes the risk assessment model and loads the patient-related data saved during the pre-operative information entry phase as static input features. These static features include clinical parameters that do not change over time or remain relatively stable during the examination, such as age, sex, body mass index, comorbidities, fasting duration, and the state of gastric contents. This step ensures that the model has an accurate understanding of the patient's baseline risk level from the beginning of the examination, providing a reliable baseline context for subsequent real-time assessment by integrating dynamic data.

[0067] Step S420: During the gastroscopy examination, intraoperative operation data from the endoscopy host and physiological parameters from the monitoring device are continuously collected at preset time intervals as dynamic input features.

[0068] In step S420, key dynamic variables reflecting the progress of the procedure and the patient's physiological response are continuously acquired to construct the time-dimensional input for risk assessment. During the gastroscopy, the system automatically collects two types of core data at preset time intervals: one type is intraoperative operational data from the endoscopy unit, such as carbon dioxide injection rate, cumulative fluid injection volume, and operation duration; the other type is physiological parameters from monitoring equipment, such as blood oxygen saturation (SpO2), heart rate, and respiratory rate. These data, as dynamic input features, can reflect changes in intragastric pressure, the intensity of operative stimulation, and the patient's physiological compensatory ability to respond to stimulation in real time. Through periodic acquisition, the system establishes a continuous time-series data stream, enabling risk assessment to capture the dynamic trend of risk evolution rather than relying solely on instantaneous snapshots.

[0069] Step S430 involves adaptively fusing static and dynamic input features using a self-attention mechanism to generate a weighted fused temporal feature representation. The self-attention mechanism dynamically allocates fusion weights based on the correlation between each feature and the current risk of backflow aspiration.

[0070] In step S430, a self-attention mechanism is used to achieve intelligent weighted fusion of static and dynamic features, improving the feature representation's ability to discriminate the current risk state. The system utilizes the self-attention mechanism to jointly process the loaded static input features and the continuously flowing dynamic input features. This mechanism automatically assigns different fusion weights based on the correlation strength between each feature item and the current risk of reflux aspiration. For example, when a rapid decrease in SpO2 is detected, the system assigns a higher attention weight to this physiological parameter; if the patient has a history of gastroparesis, the corresponding static feature weight will also be increased accordingly. This adaptive fusion method avoids the subjectivity and rigidity of manually setting fixed weights, allowing the model to focus on the most relevant risk signals and generate a more discriminative temporal feature representation.

[0071] Step S440: Input the weighted fusion time series feature representation into the risk assessment model and output the probability of backflow aspiration risk at the current moment.

[0072] In step S440, the fused high-dimensional temporal features are transformed into a quantifiable probability output of reflux aspiration risk. The system inputs the time-series feature representation, weighted and fused by a self-attention mechanism, into a pre-trained risk assessment model. This model learns the complex nonlinear relationship between features and aspiration events based on a large amount of historical case data. After internal calculation, the model outputs a value between 0 and 1, representing the probability of reflux aspiration occurring at the current moment. This probability value serves as an objective and continuous risk quantification indicator, providing a precise basis for subsequent grading judgments and intervention decisions, overcoming the limitations of traditional methods that rely solely on qualitative or binary judgments.

[0073] Step S450: Based on the probability of backflow and accidental aspiration, determine the corresponding risk level and update and display it in real time in the monitoring dashboard area 103.

[0074] In step S450, the abstract risk probability is transformed into an intuitive and operable risk level, and the information is communicated in real time through a visual interface. Based on the output regurgitation and aspiration risk probability, the system determines the corresponding risk level according to preset grading standards (such as low, medium, and high risk ranges), and updates and displays this level in real time on the monitoring dashboard area 103 of the main interface 100. This visual presentation allows the operating physician to quickly grasp the current risk situation without interpreting the original probability values. Simultaneously, the dynamic updating of the risk level ensures that the interface information is synchronized with the patient's actual condition, providing front-end support for timely identification of rising risk trends and triggering subsequent intervention suggestion windows.

[0075] In some embodiments of this application, the real-time monitoring and risk feedback interface of the monitoring dashboard area 103 is used to dynamically present the patient's risk status and key monitoring data during the examination. The top of the interface displays the patient's basic information, including name Li Si, gender female, age 48, body mass index 32 kg / m², and a three-year history of gastroparesis, providing a basic risk background for clinical practice. The left side is the real-time monitoring data display area, listing key vital signs and operational parameters during the operation, including the current carbon dioxide injection rate of 60 ml / min, which has exceeded the warning threshold of 50 ml / min and is marked as exceeding the standard; the cumulative fluid injection volume is 350 ml, which is lower than the warning value of 600 ml, indicating a normal status; the blood oxygen saturation SpO2 is 94%, which is higher than the minimum limit of 90%, indicating a normal status; the heart rate is 88 beats / min, which is within the normal range of 60-100 beats / min; and the respiratory rate is 18 breaths / min, which is within the normal range of 12-20 breaths / min.

[0076] The right-hand area displays the risk assessment results. The current risk level is medium risk, with a risk probability of 0.62. The system indicates that the risk probability has gradually increased from 0.35 to 0.62 within the past 10 minutes, showing a continuous upward trend, indicating that the risk is dynamically evolving. The system further analyzes the contribution of key risk factors, with excessively rapid CO2 injection rate contributing 33%, a history of gastroparesis contributing 25%, a patient's high BMI contributing 20%, a slight decrease in SpO2 contributing 5%, and other factors contributing 15%, clearly revealing the main driving factors of the increased risk. In addition, a postoperative risk prediction area is reserved at the bottom of the interface, indicating that there are currently no postoperative risk prediction results and that an assessment will be conducted after the surgery. This interface, through its structured layout and visual annotations, enables real-time perception, quantitative analysis, and interpretable feedback of intraoperative risks, supporting doctors to adjust their operating strategies in a timely manner and improve perioperative safety.

[0077] In some embodiments of this application, in the preoperative information entry sub-interface 200, static input features are automatically retrieved through the hospital information system and electronic medical record interface. These features include demographic information such as age, gender, height, and weight for calculating body mass index; comorbidities such as history or symptoms related to delayed gastric emptying, Parkinson's disease, stroke, gastroparesis, hiatal hernia, etc.; surgical history including previous abdominal surgery or gastrectomy; medication use covering opioid analgesics, anticholinergics, and prokinetic drugs; and imaging results such as the cross-sectional area of ​​the gastric antrum in the abdominal ultrasound report or the description of the gastric filling status in CT and MRI examinations.

[0078] Dynamic input features are collected in real time via the device interface, including carbon dioxide injection flow rate, water injection flow rate, cumulative injection volume, and injection duration output by the endoscope host; heart rate, blood oxygen saturation, and respiratory rate output by the monitor; and the actual fasting time, examination position (left lateral decubitus or supine), and anesthesia method (awake sedation or general anesthesia) manually entered by the anesthesiologist or operator. All collected data are precisely aligned according to patient identification and timestamps and uniformly stored in a structured time-series database such as InfluxDB or TimescaleDB, providing a high-quality and highly consistent data foundation for subsequent risk modeling and real-time assessment.

[0079] In some embodiments of this application, the original multi-source clinical data is cleaned, normalized, and feature-constructed to form a standardized input representation that can be recognized and processed by artificial intelligence models. During the data cleaning stage, the system automatically removes obvious outliers, such as data with oxygen saturation greater than 100%, marks unfilled items as missing and prompts the user to supplement the input, or fills in data using different conservative estimation strategies based on patient type, such as emergency or elective examination. In the feature encoding stage, continuous variables, including body mass index, total carbon dioxide injection, and total fluid injection, retain their original values ​​and undergo standardization. Categorical variables, such as examination position and anesthesia method, are converted using one-hot encoding. Time-series variables, such as the trend of oxygen saturation changes, are quantified by extracting statistical features such as slope and fluctuation amplitude.

[0080] In some embodiments of this application, the system also constructs several key derived features, among which the intragastric pressure proxy index is calculated by dividing the cumulative amount of carbon dioxide and water injected by the actual fasting time, and is used to reflect the intensity of gas and liquid load per unit time; the risk accumulation index is obtained by multiplying the standardized value of each risk factor by its corresponding weight and then summing them, with the initial weights set according to authoritative clinical guidelines such as the American Society of Anesthesiologists or the European Society of Gastrointestinal Endoscopy.

[0081] Specifically, the intragastric pressure surrogate index is a derived feature calculated by cross-referencing intraoperative dynamic variables (such as the cumulative volume of carbon dioxide and water injected) and preoperative static variables (such as the actual fasting time). Its core function is to indirectly quantify the changing trend of intragastric pressure load and reflect the impact of operative actions on gastric distension per unit time. Since real intragastric pressure is difficult to monitor continuously non-invasively, this index constructs a surrogate parameter with clear physiological significance by dividing the total injected volume by the fasting time, effectively coupling the intensity of the operation with the patient's baseline condition. When the self-attention mechanism integrates static and dynamic features, this index acts as a cross-modal bridge, guiding the model to identify clinical interaction logics such as "high gas injection volume is more dangerous in patients with short fasting periods," thereby dynamically enhancing the fusion weight of relevant features and making risk assessment more consistent with the actual pathophysiological process.

[0082] The risk accumulation index is a comprehensive score formed by weighted summation of standardized static features (such as body mass index, comorbidities, and gastric antral cross-sectional area) after assigning initial weights to different preoperative risk factors according to authoritative clinical guidelines (such as ASA and ESGE). It characterizes the patient's baseline susceptibility to reflux and aspiration. This index not only condenses multidimensional static information but also embeds prior medical knowledge, serving as a high-level semantic summary of static features within the self-attention mechanism, helping the model quickly establish a patient risk baseline. When the index is high, the system automatically increases its sensitivity to minor intraoperative physiological fluctuations (such as a slight decrease in SpO2) or procedural changes, assigning higher attention weights to corresponding dynamic features during the fusion process, reflecting the clinical principle that "high-risk patients require more cautious responses." Therefore, the risk accumulation index significantly enhances the individualization and situational adaptability of the self-attention mechanism in risk perception.

[0083] In some embodiments of this application, after the above preprocessing is completed, the system organizes the preoperative static features and intraoperative dynamic temporal features into independent feature sequences and inputs them into the self-attention mechanism module. This mechanism jointly models the static and dynamic features, dynamically assigning fusion weights based on the correlation strength between each feature and the current risk of reflux aspiration, thereby generating a weighted fused temporal feature representation. For example, when a rapid decrease in blood oxygen saturation is detected, the system automatically increases the attention weight of this physiological parameter in the fusion; if the patient has a history of gastroparesis, the corresponding static feature is also assigned a higher weight. This adaptive fusion method eliminates the subjectivity and rigidity of manually setting fixed weights, enabling the model to focus on the most relevant risk signals in real time, significantly enhancing the feature representation's ability to discriminate the current clinical risk status, and providing a more targeted contextual basis for subsequent risk probability output.

[0084] In some embodiments of this application, reference is made to Figure 5 The risk intervention suggestion window 400 includes a structured suggestion display area 401, an interpretability analysis area 402, a suggestion adoption control 403, and a suggestion rejection control 404. In step S140, during the gastroscopy examination, when the risk assessment results and / or monitoring data meet the warning rules, the risk intervention suggestion window 400 automatically pops up, and pushes clinical intervention suggestions and receives doctor feedback, including the following steps.

[0085] In step S510, in response to the judgment result that the risk triggering conditions are met, the risk intervention suggestion window 400 is activated, and the current risk level and the corresponding structured clinical intervention suggestions are overlaid on the main interface 100.

[0086] In step S510, when the risk reaches a preset threshold, the human-computer interaction intervention mechanism is activated promptly to ensure that clinical staff can be informed of the risk status and obtain response guidance immediately. When the system determines that the current risk assessment result and / or monitoring data meet the pre-configured early warning rules, the risk intervention suggestion window 400 is immediately activated, and the current risk level and the corresponding structured clinical intervention suggestions are displayed on the main interface 100 in an overlay manner. This step achieves a seamless connection from risk identification to intervention prompts, breaks operational inertia through a forced pop-up window, effectively prevents risk signals from being ignored, buys doctors critical decision-making time, and is an important trigger point for ensuring patient safety.

[0087] Step S520: In the structured suggestion display area 401, a list of intervention measures matching the current risk scenario is displayed. These intervention measures include at least one of the following: suspending CO2 infusion, reducing fluid volume, adjusting patient position, or enhancing airway monitoring.

[0088] In step S520, specific, actionable clinical operational guidance is provided to physicians, closely aligned with the current risk scenario. In the structured suggestion display area 401 of the risk intervention suggestion window 400, the system lists one or more targeted intervention measures, including at least one of suspending carbon dioxide infusion, reducing fluid injection volume, adjusting patient position, or enhancing airway monitoring. Each suggestion is automatically generated based on the currently triggered risk factor; for example, if rapid CO2 injection leads to increased intragastric pressure, suspending infusion is the preferred recommendation; if SpO2 decreases significantly, enhanced airway monitoring is emphasized. This step transforms abstract risks into clear operational instructions, significantly reducing the cognitive load of clinical decision-making and improving the timeliness and standardization of interventions.

[0089] In step S530, in the interpretability analysis area 402, the contribution of each input feature to the current risk assessment result is visualized to generate an interpretability analysis report.

[0090] In step S530, the transparency and credibility of risk assessment results are enhanced, supporting physicians to make rational judgments rather than blindly relying on system output. In the interpretability analysis area 402, the system presents the contribution of each input feature to the current risk assessment result through visualization, such as bar charts, heatmaps, or weighted rankings, showing which preoperative factors or intraoperative parameters have the greatest impact on high-risk judgments. This interpretable analysis report enables physicians to understand why the system issued an alert, thereby judging the rationality of the recommendations and avoiding trust gaps or misoperations caused by black-box models. It also provides a basis for teaching and quality control, demonstrating the auditability and clinical friendliness of AI-assisted decision-making.

[0091] In step S540, in response to the trigger command of the recommendation adoption control 403, the doctor's confirmation of the implementation of the intervention recommendation is recorded, and the operation is associated with the current risk event log.

[0092] In step S540, the doctor's positive response to the system's suggestions is recorded, forming a closed-loop intervention evidence chain. When a doctor deems the intervention suggestion pushed by the system reasonable and decides to adopt it, they can confirm by clicking the suggestion adoption control 403. The system then records this operation and stores it in the risk event log, associating it with information such as the timestamp of the current risk event, risk level, and specific suggestion content. This record is not only used for subsequent quality retrospectives and operational compliance reviews but also provides positive sample data for model validation, indicating that the intervention measure is clinically recognized in a specific risk context, which helps to optimize the suggestion generation strategy in the future.

[0093] In step S550, in response to the trigger command of the suggestion rejection control 404, a remarks input window pops up, receives the rejection reason entered by the doctor, and saves the rejection operation and reason to the feedback record for subsequent iterative optimization of the risk assessment model.

[0094] In step S550, the system captures doctors' negative feedback on system suggestions and their reasons, providing high-quality negative samples and semantic context for continuous model learning. When a doctor believes the current suggestion is inapplicable or biased, the suggestion rejection control 404 can be triggered. The system then pops up a remarks input window, requiring the doctor to enter the specific reason for rejection, such as the patient's special condition, the inability to interrupt the operation, or interference with monitoring data. This rejection operation, along with the entered reason, will be completely saved in the feedback record. These semantically labeled negative cases are extremely valuable for identifying model blind spots, correcting feature weights, and improving suggestion accuracy. They are a key data source for achieving iterative optimization and clinical adaptation of the risk assessment model.

[0095] In some embodiments of this application, when the system determines that the patient is in a high-risk state, i.e., the risk probability P ≥ 0.7 or SpO2 < 90% for 15 seconds, the system immediately triggers a high-risk response procedure and automatically generates clinical recommendations containing multiple key operational instructions: including immediately checking and pausing gas and water injection, closely monitoring and calling for support and requesting instructions from superiors; turning the patient into a left lateral decubitus position with the head lowered by 30 degrees to facilitate drainage; initiating high-flow oxygen therapy; strengthening suction preparation; considering placing a nasogastric tube for decompression; and, if necessary, waking the patient to restore protective reflexes.

[0096] For medium-risk situations (risk probability P < 0.7, and the total volume of CO2 and water injected > 600 ml), the system recommends reducing the inhalation rate to 50 ml per minute, avoiding frequent head elevation, and strengthening the monitoring of carbon dioxide waveform, airway pressure, and blood oxygen saturation to prevent further risk escalation. When the risk level is low (risk probability P < 0.3), the system indicates that the procedure can continue as usual. All recommendations are generated and matched from a pre-set rule base based on the risk level, ensuring that the intervention measures are clinically reasonable, timely, and feasible, forming a closed-loop management mechanism from risk identification to precise intervention.

[0097] In some embodiments of this application, reference is made to Figure 5 In the structured suggestion display area 401, the system automatically generates and displays five specific intervention measures based on the current risk status, presented in a list format: The first suggestion is to immediately adjust the carbon dioxide injection rate to 50 ml / min or less; the second suggestion is to suspend the water injection operation until the risk level falls back to the low-risk range; the third suggestion is to adjust the patient's position to a left lateral decubitus position with the head lowered at 30 degrees, avoiding supine position operations; the fourth suggestion is to closely observe whether the patient experiences prodromal symptoms such as nausea and vomiting, and to require nurses to prepare antiemetic medication in advance, and to notify the physician for medication intervention if necessary; the fifth suggestion is to continuously monitor changes in heart rate and respiratory rate, and to take immediate action if the heart rate exceeds 100 beats per minute. Each suggestion is followed by a suggestion acceptance control 403 and a suggestion rejection control 404, allowing doctors to confirm or reject suggestions one by one, and providing "one-click acceptance" and "one-click rejection" functions to improve interaction efficiency.

[0098] Below the suggestions is an interpretability analysis area 402, which details the contribution of each risk factor to the current risk score: Excessive CO2 injection rate accounts for 33%, with the current rate of 60 ml / min exceeding the threshold of 50 ml / min; history of gastroparesis accounts for 25%, due to prolonged delayed gastric emptying leading to residual gastric contents; excessive BMI accounts for 20%, as obesity increases intra-abdominal pressure and the risk of reflux; a slight decrease in SpO2 accounts for 5%, from 97% to 94%; other factors account for 15%, including prolonged operation time and depth of anesthesia. This interface, through the combination of structured suggestions and transparent explanations, enhances doctors' trust in and willingness to adopt AI decisions, achieving effective synergy between intelligent assistance and clinical autonomous judgment.

[0099] In some embodiments of this application, reference is made to Figure 6 The risk log sub-interface 500 includes a risk event timeline control 501, an intervention record details control 502, a doctor feedback viewing control 503, and a model optimization status indicator control 504. In step S150, the risk log sub-interface 500 supports retrospective examination of risk assessment records, intervention interaction information, and doctor feedback, including the following steps.

[0100] Step S610: In response to the trigger command of the risk event timeline control 501, display the risk event nodes arranged in chronological order, with each node associated with the real-time risk probability, risk level and triggered warning rule at the corresponding time.

[0101] In step S610, the dynamic trajectory of risk evolution during gastroscopy is reconstructed in a panoramic timeline format, providing a clear temporal framework for post-operative review. When the user triggers the risk event timeline control 501, the system displays risk event nodes arranged chronologically in the risk log sub-interface 500. Each node is associated with and records the real-time risk probability output by the risk assessment model at that moment, the determined risk level, and the specific triggering warning rule. This structured timeline presentation allows doctors or quality control personnel to intuitively trace when the risk increased, which rules were activated, and at which operational stage the risk peak occurred, thereby accurately identifying high-risk operational steps or patient sensitivity windows, which forms the basis for conducting root cause analysis and process improvement.

[0102] In step S620, in response to the trigger command of the intervention record details control 502, the structured clinical intervention suggestion content and push time corresponding to the selected risk event node are displayed.

[0103] In step S620, the specific risk event is precisely linked to the specific intervention measures pushed by the system at that time, ensuring that the intervention is traceable and verifiable. When the user triggers the intervention record details control 502, the system displays the structured clinical intervention suggestions corresponding to the selected risk event node and their push time. These suggestions include specific operational items such as pausing CO2 injection and adjusting body position, and clearly indicate the precise time when the system issued the suggestion. This step allows the reviewer to determine whether the intervention suggestions are timely and whether they match the risk characteristics, and also provides direct evidence for assessing the clinical rationale of the suggestions, enhancing the transparency and auditability of the risk management process.

[0104] In step S630, in response to the trigger command of the doctor feedback viewing control 503, the doctor's operation record for the intervention suggestion is displayed. The operation record includes the acceptance confirmation or rejection remarks and the reasons for input.

[0105] In step S630, the doctor's subjective judgment and decision-making basis regarding the system's suggestions are fully preserved, forming genuine evidence of human-computer interaction. Responding to the trigger command of the doctor's feedback viewing control 503, the system displays the doctor's operation record for the corresponding intervention suggestion. This record clearly indicates whether the doctor performed an acceptance confirmation or chose to reject it, along with the specific reasons entered. This feedback not only reflects the doctor's acceptance of the suggestion but also includes key contextual information such as the patient's special condition and the uninterrupted nature of the operation. This step ensures that clinical professional judgment is effectively captured by the system, avoiding superficial human-computer interaction and providing high-quality, semantically rich feedback data for subsequent model optimization.

[0106] In step S640, in response to the trigger command of the model optimization status indicator control 504, the progress status or version update information of the iterative optimization of the risk assessment model based on the feedback records accumulated in this inspection is displayed.

[0107] In step S640, clinical practice feedback is transformed into visible results of model evolution, establishing a positive closed loop from use to optimization. When the user triggers the model optimization status indicator control 504, the system displays the progress status or version update information of the iterative optimization of the risk assessment model based on the doctor's feedback records accumulated during this gastroscopy. For example, it can show whether the model has been retrained based on the rejection reasons in this case, the current running version number, the optimization task queuing status, or the estimated time of the next update. This step allows the clinical team to perceive that their feedback is actually used for system improvement, enhancing their willingness to use it. It also demonstrates that the risk management system constructed in this application has the ability to continuously learn and self-evolve, rather than being a statically deployed, one-off tool.

[0108] In some embodiments of this application, the risk assessment model adopts an edge-cloud collaborative architecture, including a large global model deployed in the cloud and a lightweight model deployed on an edge computing device at the endoscopy site. The large global model is trained based on multi-center historical case data, while the lightweight model is generated from the large global model through model compression or knowledge distillation and deployed on the edge computing device to support local real-time inference.

[0109] Specifically, an edge-cloud collaborative risk assessment architecture that balances high accuracy and low latency is constructed to meet the comprehensive requirements of real-time performance, privacy, and model performance in gastroscopy reflux aspiration risk management. By dividing the risk assessment model into a large global model deployed in the cloud and a lightweight model deployed on edge computing devices at the gastroscopy site, the system achieves a reasonable division of computing resources and data processing capabilities.

[0110] The global large-scale model is trained using historical case data accumulated from multiple centers, possessing strong generalization capabilities and complex feature learning abilities, continuously extracting risk patterns from massive amounts of clinical practice. The lightweight model, on the other hand, extracts core decision-making logic from the global large-scale model through model compression or knowledge distillation techniques, significantly reducing computational complexity and resource consumption while retaining key predictive capabilities, thus adapting to the hardware limitations of edge devices. This lightweight model performs real-time inference locally, responding quickly to dynamic changes during surgery without relying on network transmission, ensuring low latency and high availability for risk assessment, while avoiding frequent uploads of sensitive patient physiological data to the cloud, enhancing data privacy protection. This architecture leverages the advantages of cloud-based big data training while meeting the requirements of immediate response and reliable operation in clinical settings, providing robust technical support for intelligent risk warning.

[0111] In some embodiments of this application, the method involves the continuous co-evolution of a risk assessment model through the following steps.

[0112] Step S710: After completing the gastroscopy on the edge computing device, the structured feedback data from this examination is encrypted and uploaded to the cloud.

[0113] Step S710 enables the secure backflow of real-world clinical feedback data, providing high-quality training material for model optimization. After the gastroscopy is completed on the edge computing device, the system encrypts the structured feedback data generated during the examination and uploads it to the cloud. This data includes key interactive information such as the doctor's adoption or rejection of intervention suggestions, reasons for rejection, risk event records, and the actual implementation of intervention measures, all of which have been standardized and anonymized. Encrypted uploading ensures both patient privacy and medical data security, while also ensuring that valuable clinical experience accumulated at the edge can be effectively aggregated to the central platform, laying a data foundation for the iteration of the global model.

[0114] Step S720: In the cloud, the global large model is periodically retrained based on the newly collected feedback data.

[0115] In step S720, continuously collected multi-source feedback data drives the knowledge update and performance improvement of the global large-scale model. After receiving structured feedback data uploaded by various edge devices in the cloud, the system incorporates it into the training data pool and, combined with existing multi-center historical cases, periodically retrains the global large-scale model. This retraining process enables the model to continuously learn new risk patterns, correct existing judgment biases, and adapt to the operational habits of different medical institutions and patient group characteristics, thereby enhancing its generalization ability, accuracy, and clinical applicability. This step is the core link in realizing the intelligent evolution of the model, ensuring that the global large-scale model always reflects the latest clinical practice knowledge.

[0116] In step S730, the updated global large model is compressed or distilled again into a new lightweight model and distributed to each edge computing device to achieve continuous collaborative evolution of the risk assessment model.

[0117] In step S730, the optimized model capabilities from the cloud are efficiently deployed back to the edge, completing the closed-loop transfer from central learning to local application. After the global large model is retrained, the system uses model compression or knowledge distillation techniques to transform it into a new lightweight model suitable for edge computing devices. This new model maintains inference accuracy while meeting the stringent requirements of edge devices for computing resources, memory usage, and response speed. Subsequently, the system automatically distributes this lightweight model to each edge computing device deployed on-site, replacing the old version, thereby achieving an overall upgrade of risk assessment capabilities. Through this mechanism, all terminal devices can synchronously obtain the latest optimization results, ensuring that the entire system maintains a consistent and continuously evolving level of intelligence in a distributed environment, truly realizing the dynamic evolution of edge-cloud collaboration.

[0118] In some embodiments of this application, the monitoring dashboard area 103 is also used to display the risk assessment results of postoperative reflux aspiration predicted by the risk assessment model, specifically including the following steps.

[0119] In step S810, in response to the end command of the gastroscopy examination, based on the intraoperative operation data accumulated during the examination, the trend of physiological parameter changes and the final gastric state, the risk assessment model is used to predict the probability of delayed gastric emptying and the recovery time of airway protection ability during the postoperative recovery period.

[0120] In step S810, the multidimensional dynamic data accumulated during the procedure is extended to postoperative risk prediction, enabling a prospective assessment of the patient's safety status during the recovery period. Upon receiving the end-of-gastroscopy command, the risk assessment model immediately activates the postoperative extrapolation module. This module comprehensively analyzes the accumulated intraoperative data (such as total CO2 injection, fluid injection volume, and procedure duration), physiological parameter trends (such as SpO2 fluctuations and heart rate variability), and the final gastric state (such as antral cross-sectional area or Perlas classification). Based on this analysis, the probability of delayed gastric emptying and the time required for the recovery of airway protective reflexes (such as swallowing and coughing reflexes) during the postoperative recovery phase are calculated. This step overcomes the limitations of traditional methods that only focus on intraoperative risks, extending the risk management chain to the critical postoperative window and providing a scientific basis for subsequent nursing decisions.

[0121] Step S820: Generate a postoperative reflux and aspiration risk score based on the probability of delayed gastric emptying and the recovery time of airway protection capacity.

[0122] In step S820, complex physiological and operational variables are transformed into unified and comparable quantitative indicators, facilitating rapid clinical identification of high-risk patients. Based on the two core prediction results output in step S810—the probability of delayed gastric emptying and the recovery time of airway protection ability—the system generates a comprehensive postoperative reflux and aspiration risk score through preset mapping rules or weighted algorithms. This score reflects the overall probability of aspiration during the recovery period in numerical form, considering both the physical risk of residual gastric contents and the functional risk of incomplete recovery of neurological reflexes. This achieves a structured and standardized expression of postoperative risk, providing objective support for hierarchical management and intervention prioritization.

[0123] In step S830, the postoperative reflux and aspiration risk score is displayed in the monitoring dashboard area 103 as an independent postoperative risk card, and the high-risk time period window is marked.

[0124] In step S830, postoperative risk information is conveyed to medical staff in a clear and prominent manner to ensure that no critical warnings are missed. The system displays the generated postoperative regurgitation and aspiration risk score as a separate postoperative risk card in monitoring dashboard area 103, clearly marking high-risk time periods on the card, such as "30 minutes to 2 hours postoperatively is a high-risk period." This card-based design, distinct from the intraoperative real-time data stream, creates a visual focus, allowing anesthesia nurses, recovery room doctors, or shift handover personnel to quickly obtain an overview of the patient's postoperative risk and time sensitivity, effectively supporting the rational allocation of monitoring resources during the recovery period and the accurate transmission of handover information.

[0125] Step S840: When the postoperative reflux and aspiration risk score exceeds the preset threshold, postoperative care suggestions are pushed to the main interface 100. The postoperative care suggestions include at least one of the following: maintaining a lateral decubitus position, delaying drinking and eating, strengthening SpO2 monitoring, or extending the PACU observation time.

[0126] In step S840, the postoperative risk score is directly translated into specific and actionable clinical nursing measures, achieving a closed loop from risk identification to intervention implementation. When the system determines that the postoperative regurgitation and aspiration risk score exceeds a preset threshold, it immediately pushes postoperative nursing suggestions on the main interface. These suggestions include maintaining a lateral decubitus position to reduce the possibility of regurgitation, delaying drinking and eating until airway protection is restored, strengthening SpO2 monitoring to detect oxygenation abnormalities early, or extending the observation time in the post-anesthesia care unit (PACU) at least one of the following:

[0127] In some embodiments of this application, the main interface 100 is also adapted to a mobile terminal, specifically including the following:

[0128] (1) When the risk intervention suggestion window 400 is triggered, or when the postoperative reflux aspiration risk score exceeds the preset threshold, a card-style notification is pushed to the mobile terminal. This breaks through the spatial limitations of fixed workstations and ensures that key risk information can reach clinicians in a timely manner, regardless of whether they are in the operating room. When the system triggers the risk intervention suggestion window 400 during gastroscopy, or when the postoperative reflux aspiration risk score is determined to exceed the preset threshold after the examination, a card-style notification is automatically pushed to the bound mobile terminal. The notification presents the core prompts of the risk event in a concise and eye-catching form, such as "High risk: It is recommended to suspend insufflation" or "High risk after surgery: Observation needs to be extended", so that doctors can be informed of the emergency or important risk status as soon as possible, even when they are making rounds in the ward, treating other patients, or on their way to shift handover, which significantly improves the timeliness of response and the continuity of management.

[0129] (2) In response to the click operation of the card-style notification, a risk snapshot interface is displayed on the mobile terminal. The risk snapshot interface includes snapshots of key parameters such as CO2 injection rate or cumulative injection volume, SpO2 blood oxygen saturation value, and the current risk level. Therefore, it can provide necessary contextual information support for rapid decision-making in mobile scenarios and avoid misjudgment due to insufficient information. When the doctor clicks on the pushed card-style notification, the mobile terminal immediately displays the risk snapshot interface, which focuses on presenting snapshots of several key parameters most relevant to the current risk, including CO2 injection rate or cumulative injection volume, SpO2 blood oxygen saturation value, and the current risk level. These parameters are filtered and condensed, reflecting both the intensity of the operational stimulus and the patient's physiological response, which is sufficient to support the doctor to make a preliminary judgment on the authenticity and urgency of the risk in a mobile state, and to grasp the core situation without returning to the main console, thereby improving the accuracy and efficiency of remote intervention.

[0130] (3) Configure one-click confirmation and one-click rejection controls in the risk snapshot interface to support doctors' rapid feedback; respond to the trigger command of the one-click rejection control, enable the voice input function, receive the rejection remarks input by the doctor, realize an efficient and low-burden human-computer feedback loop on the mobile terminal, and especially adapt to the operating habits in the high-intensity clinical work environment. The risk snapshot interface is configured with one-click confirmation and one-click rejection controls. Doctors only need to click once to complete the operation of accepting or rejecting the system's suggestions, which greatly simplifies the interaction process. In particular, when the doctor triggers the one-click rejection control, the system automatically enables the voice input function, allowing the doctor to directly state the rejection reason, such as "the patient just vomited and the stomach has been emptied" or "SpO2 dropped due to the probe being loose", without having to type manually. This design fully considers the practical constraints of the inconvenience of input for medical staff when they are moving or wearing gloves. It efficiently captures high-quality semantic feedback through voice, which not only ensures the user experience, but also continuously accumulates real and structured negative sample data for model optimization.

[0131] Secondly, refer to Figure 7 This application provides an interactive system for managing the risk of reflux aspiration during gastroscopy, including a main interface module, a preoperative information entry module, a risk warning configuration module, a monitoring dashboard module, and a risk log viewing module.

[0132] The main interface module is configured as follows: it displays the main interface for managing the risk of gastroscopy reflux aspiration. The main interface includes preoperative information entry controls, risk warning configuration controls, monitoring dashboard area, and risk log viewing controls.

[0133] The preoperative information entry module is configured to: respond to the trigger command of the preoperative information entry control, display the preoperative information entry sub-interface, and enter the patient's relevant preoperative information.

[0134] The risk warning configuration module is configured to: respond to the trigger command of the risk warning configuration control, display the risk warning configuration sub-interface, provide intelligently recommended default warning rules, and support the adjustment of the default warning rules based on preoperative information to complete the configuration of warning rules for the risk of reflux aspiration.

[0135] The monitoring dashboard module is configured to display real-time monitoring data during gastroscopy, and generate real-time risk assessment results for reflux aspiration based on preoperative information and monitoring data using a risk assessment model. When the risk assessment results and / or monitoring data meet the warning rules, a risk intervention suggestion window automatically pops up on the main interface, pushing clinical intervention suggestions and receiving feedback from doctors.

[0136] The risk log viewing module is configured to display a risk log sub-interface in response to a trigger command on the risk log viewing control. It supports retrospective analysis of risk assessment records, intervention interaction information, and doctor feedback during the examination process to support continuous optimization of the risk assessment model.

[0137] In some embodiments of this application, the system is deployed on a local server or private cloud platform in a hospital endoscopy center. It is interconnected with the hospital information system, electronic medical records, endoscopy host, monitor and ultrasound equipment through a standard medical interface. The system front end supports access via web browser, tablet and mobile application. The back end is composed of a data engine, AI inference engine and interactive service engine, forming a complete technical system that integrates data integration, intelligent analysis and human-computer interaction.

[0138] In summary, the interactive method and system for managing the risk of reflux aspiration during gastroscopy provided in this application have the following technical effects.

[0139] This method, by constructing an intelligent interactive platform covering the entire preoperative, intraoperative, and postoperative process, achieves a paradigm shift from static experience-based judgment to dynamic data-driven risk management. The main interface integrates four core modules: preoperative information entry, early warning rule configuration, real-time monitoring, and risk log retrospective analysis. This unified operation entry point eliminates data silos, significantly improving clinical efficiency and information completeness. The system intelligently recommends personalized preoperative information and supports custom early warning rules. Combined with multi-source dynamic monitoring data during surgery, it utilizes a risk assessment model incorporating a self-attention mechanism to achieve high-precision, interpretable real-time risk probability output. When a risk is triggered, it automatically pushes structured intervention suggestions and simultaneously receives feedback from physicians regarding acceptance or rejection, forming a human-machine collaborative decision-making closed loop. In the postoperative stage, risk prediction is further extended to the recovery period, generating a postoperative reflux and aspiration risk score and indicating high-risk periods and nursing measures, expanding the boundaries of safety management. The risk log sub-interface supports full-cycle event retrospective analysis and feedback accumulation, providing high-quality training data for continuous model optimization.

[0140] Furthermore, the system adopts an edge-cloud collaborative architecture. Lightweight models at the edge ensure low-latency local inference, while the global model in the cloud is periodically retrained using multi-center data. Continuous collaborative evolution of the model is achieved through encrypted uploads, model distillation, and version distribution mechanisms. Simultaneously, the main interface is adapted for mobile terminals, supporting card-style risk notifications, snapshot display of key parameters, and voice input rejection feedback, ensuring timely response and efficient interaction for doctors in any scenario. Overall, this method effectively addresses the core shortcomings of existing technologies, such as single assessment dimensions, scattered data, lack of dynamic updates, lack of interpretability, and human-machine disconnect. It significantly improves the accuracy of reflux aspiration risk identification, the timeliness of intervention, the scientific nature of decision-making, and the system's clinical applicability and adaptability during gastroscopy.

[0141] It should be noted that in all specific embodiments of this application, all data processing activities related to user identity or personal characteristics, such as user information, user behavior data, historical data, and location information, will be conducted in accordance with the principles of legality, legitimacy, and necessity. All data collection, use, storage, and processing will be subject to compliance with applicable national and regional laws, regulations, and industry standards, and informed consent from users will be obtained in a clear and explicit manner before processing. For the processing of sensitive personal information, separate consent from users will be obtained through prominent means such as pop-up prompts and independent confirmation pages. If any processing conflicts with laws and regulations, the laws and regulations will prevail, and necessary data processing will only be carried out within the scope permitted by laws and regulations, ensuring that all data-based applications, analyses, and technical implementations are conducted within the scope permitted by laws and regulations.

[0142] In some alternative embodiments, the functions / operations mentioned in the block diagrams may not occur in the order shown in the operation diagrams. For example, depending on the functions / operations involved, two consecutively shown blocks may actually be executed substantially simultaneously, or the blocks may sometimes be executed in reverse order. Furthermore, the embodiments presented and described in the flowcharts of this application are provided by way of example to provide a more comprehensive understanding of the technology. The disclosed methods are not limited to the operations and logic flows presented herein. Alternative embodiments are contemplated in which the order of various operations is changed and sub-operations described as part of a larger operation are executed independently.

[0143] Furthermore, although this application is described in the context of functional modules, it should be understood that, unless otherwise stated, one or more of the functions and / or features may be integrated into a single physical device and / or software module, or one or more functions and / or features may be implemented in a separate physical device or software module. It is also understood that a detailed discussion of the actual implementation of each module is unnecessary for understanding this application. Rather, given the properties, functions, and internal relationships of the various functional modules in the apparatus disclosed herein, the actual implementation of the module will be understood within the scope of ordinary skill of an engineer. Therefore, those skilled in the art can implement the application set forth in the claims using ordinary skill. It is also understood that the specific concepts disclosed are merely illustrative and are not intended to limit the scope of this application, which is determined by the full scope of the appended claims and their equivalents.

[0144] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several programs to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0145] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequential list of executable programs for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, a program execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can retrieve and execute a program from or in conjunction with such a program execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can mean any means that can contain, store, communicate, propagate, or transmit a program for use by or in conjunction with a program execution system, apparatus, or device.

[0146] More specific examples (a non-exhaustive list) of computer-readable media include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Additionally, computer-readable media can even be paper or other suitable media on which programs can be printed, for example, by optically scanning the paper or other media, then editing, interpreting, or, if necessary, processing it in a suitable manner to obtain the program electronically, and then storing it in computer memory.

[0147] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable program execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0148] In the foregoing description of this specification, the reference to terms such as "one embodiment / implementation," "another embodiment / implementation," or "certain embodiments / implementations," etc., indicates that a specific feature, structure, material, or characteristic described in connection with an embodiment or example is included in an embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0149] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

[0150] The above is a detailed description of the preferred embodiments of the present invention. However, the present invention is not limited to the embodiments. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of the present invention. All such equivalent modifications or substitutions are included within the scope defined by the claims of the present invention.

Claims

1. An interactive method for managing the risk of reflux aspiration during gastroscopy, characterized in that, Includes the following steps: The main interface for managing the risk of gastroscopy reflux aspiration includes preoperative information entry controls, risk warning configuration controls, a monitoring dashboard area, and risk log viewing controls. In response to a trigger command on the preoperative information entry control, a preoperative information entry sub-interface is displayed to enter relevant preoperative information of the patient; In response to the trigger command of the risk warning configuration control, a risk warning configuration sub-interface is displayed, providing intelligently recommended default warning rules, and supporting the adjustment of the default warning rules based on the preoperative information to complete the configuration of warning rules for the risk of reflux aspiration. During the gastroscopy, the monitoring data is displayed in real time through the monitoring dashboard area, and based on the preoperative information and the monitoring data, a real-time risk assessment result of reflux aspiration is generated using a risk assessment model. When the risk assessment results and / or the monitoring data meet the early warning rules, a risk intervention suggestion window will automatically pop up on the main interface, push clinical intervention suggestions and receive doctor feedback. In response to the trigger command of the risk log viewing control, the risk log sub-interface is displayed, which supports retrospective examination of risk assessment records, intervention interaction information and doctor feedback, so as to support the continuous optimization of the risk assessment model.

2. The interactive method for managing the risk of reflux aspiration during gastroscopy according to claim 1, characterized in that, The preoperative information entry sub-interface includes a patient information entry control, a medical history and comorbidity configuration control, a fasting time setting control, a gastric contents status entry control, and a preoperative information saving control. On the preoperative information entry sub-interface, enter the patient's relevant preoperative information, including the following steps: In response to a trigger command on the patient information entry control, a patient information input window is displayed, and the patient information entered includes the patient's age, gender, weight, height, and corresponding body mass index; In response to a trigger command to the medical history and comorbidity configuration control, a comorbidity selection window is displayed, and medical history items related to the risk of reflux and aspiration are selected, including at least one of diabetes, gastroparesis, hiatal hernia, or a history of aspiration. In response to the trigger command of the fasting time setting control, the system inputs the time point of the patient's last meal or drink, and automatically calculates the actual fasting duration; In response to a trigger command to the gastric contents status recording control, gastric contents status information is recorded, including at least one of the following: gastric antrum cross-sectional area CSA value, Perlas grade, or "not evaluated" label; In response to a trigger command to the preoperative information saving control, the recorded patient information, medical history and comorbidities, fasting time and gastric contents status information are saved, and a dataset of patient-related preoperative information is generated.

3. The interactive method for managing the risk of reflux aspiration during gastroscopy according to claim 1, characterized in that, The risk warning configuration sub-interface includes an intelligent recommendation control, a threshold setting control, an alarm method configuration control, and a rule saving control; In the risk warning configuration sub-interface, configuring the warning rules includes the following steps: In response to the trigger command of the intelligent recommendation control, a default warning rule matching the patient's risk characteristics is automatically loaded based on the preoperative information; In response to a trigger command on the threshold setting control, a threshold setting window is displayed, allowing modification of one or more conditions in the default warning rules. The conditions include a risk probability threshold, an intraoperative operation parameter threshold, or a physiological parameter change threshold. The intraoperative operation parameter threshold includes a carbon dioxide injection rate threshold or a cumulative fluid injection volume threshold, and the physiological parameter change threshold includes a threshold for the decrease in blood oxygen saturation per unit time. In response to a trigger command to the alarm mode configuration control, a prompting mode is configured when a risk is triggered, wherein the prompting mode includes at least one of pop-up reminder, sound alarm, or interface highlighting; In response to the trigger command of the rule saving control, the finally confirmed early warning rule and the corresponding alarm method are saved.

4. The interactive method for managing the risk of reflux aspiration during gastroscopy according to claim 1, characterized in that, The process of generating a real-time risk assessment result for reflux aspiration based on the preoperative information and the monitoring data using a risk assessment model includes the following steps: In response to the gastroscopy start command, the risk assessment model is initialized and the preoperative information is loaded as static input features; During the gastroscopy, intraoperative operation data from the endoscope host and physiological parameters from the monitoring equipment are continuously collected at preset time intervals as dynamic input features; The static input features and the dynamic input features are adaptively fused using a self-attention mechanism to generate a weighted fused temporal feature representation; wherein, the self-attention mechanism dynamically allocates fusion weights based on the correlation between each feature and the current risk of backflow aspiration. The weighted and fused temporal feature representation is input into the risk assessment model, and the probability of backflow and mis-absorption risk at the current moment is output. Based on the probability of backflow and accidental aspiration, the corresponding risk level is determined and updated and displayed in real time in the monitoring dashboard area.

5. The interactive method for managing the risk of reflux aspiration during gastroscopy according to claim 1, characterized in that, The risk intervention suggestion window includes a structured suggestion display area, an interpretability analysis area, a suggestion adoption control, and a suggestion rejection control; During a gastroscopy, when the risk assessment results and / or the monitoring data meet the warning rules, a risk intervention suggestion window automatically pops up, and clinical intervention suggestions are pushed and doctor feedback is received, including the following steps: In response to the determination that the risk triggering conditions are met, the risk intervention suggestion window is activated, and the current risk level and the corresponding structured clinical intervention suggestions are displayed on the main interface. In the structured suggestion display area, a list of intervention measures matching the current risk scenario is displayed. The intervention measures include at least one of suspending carbon dioxide injection, reducing the amount of fluid injected, adjusting the patient's position, or strengthening airway monitoring. In the interpretability analysis area, the contribution of each input feature to the current risk assessment result is visualized to generate an interpretability analysis report; In response to the trigger command of the suggestion adoption control, the doctor's confirmation of the implementation of the intervention suggestion is recorded, and the operation is associated with the current risk event log; In response to the trigger command of the suggestion rejection control, a remarks input window pops up, receives the rejection reason entered by the doctor, and saves the rejection operation and reason to the feedback record for subsequent iterative optimization of the risk assessment model.

6. The interactive method for managing the risk of reflux aspiration during gastroscopy according to claim 1, characterized in that, The risk log sub-interface includes a risk event timeline control, an intervention record details control, a doctor feedback viewing control, and a model optimization status indicator control; The risk log sub-interface supports retrospective review of risk assessment records, intervention interaction information, and doctor feedback during the examination process, including the following steps: In response to the trigger command of the risk event timeline control, risk event nodes are displayed in chronological order, with each node associated with the real-time risk probability, risk level, and triggering warning rules at the corresponding moment; In response to the trigger command of the intervention record details control, the structured clinical intervention suggestion content and push time corresponding to the selected risk event node are displayed; In response to a trigger command to view the doctor's feedback control, the doctor's operation record for the intervention suggestion is displayed, including acceptance confirmation or rejection notes and the reasons for input; In response to the trigger command of the model optimization status indicator control, the progress status or version update information of the iterative optimization of the risk assessment model based on the feedback records accumulated in this inspection is displayed.

7. The interactive method for managing the risk of reflux aspiration during gastroscopy according to claim 1, characterized in that, The risk assessment model adopts an edge-cloud collaborative architecture, including a global large model deployed in the cloud and a lightweight model deployed on the edge computing device at the gastroscopy examination site. The global large model is trained based on multi-center historical case data, and the lightweight model is generated from the global large model through model compression or knowledge distillation and deployed on the edge computing device to support local real-time inference.

8. The interactive method for managing the risk of reflux aspiration during gastroscopy according to claim 7, characterized in that, It also includes the following steps: After the gastroscopy is completed on the edge computing device, the structured feedback data from the examination is encrypted and uploaded to the cloud. In the cloud, the global large model is periodically retrained based on newly collected feedback data; The updated global model is compressed or distilled again into a new lightweight model and distributed to various edge computing devices to achieve the continuous collaborative evolution of the risk assessment model.

9. The interactive method for managing the risk of reflux aspiration during gastroscopy according to claim 1, characterized in that, The monitoring dashboard area is also used to display the risk assessment results of postoperative reflux aspiration predicted by the risk assessment model, specifically including: In response to the end command of the gastroscopy, based on the intraoperative operation data accumulated during the examination, the trend of physiological parameter changes and the final gastric state, the risk assessment model is used to infer the probability of delayed gastric emptying and the recovery time of airway protection ability during the postoperative recovery period. Based on the gastric emptying delay probability and the airway protection recovery time, a postoperative reflux aspiration risk score is generated. The postoperative reflux and aspiration risk score is displayed as an independent postoperative risk card in the monitoring dashboard area, and high-risk time periods are marked. When the postoperative reflux and aspiration risk score exceeds a preset threshold, postoperative care suggestions are pushed to the main interface. The postoperative care suggestions include at least one of the following: maintaining a lateral decubitus position, delaying drinking and eating, strengthening blood oxygen saturation monitoring, or extending the PACU observation time.

10. An interactive system for managing the risk of reflux aspiration during gastroscopy, characterized in that, Includes the following modules: The main interface module is configured to display the main interface for managing the risk of gastroscopy reflux aspiration. The main interface includes preoperative information entry controls, risk warning configuration controls, a monitoring dashboard area, and risk log viewing controls. The preoperative information entry module is configured to: in response to a trigger command on the preoperative information entry control, display a preoperative information entry sub-interface and enter relevant preoperative information of the patient; The risk warning configuration module is configured to: in response to the trigger command of the risk warning configuration control, display the risk warning configuration sub-interface, provide intelligently recommended default warning rules, and support the adjustment of the default warning rules based on the preoperative information to complete the configuration of the warning rules for the risk of reflux aspiration. The monitoring dashboard module is configured to: display monitoring data in real time through the monitoring dashboard area during the gastroscopy examination, and generate a real-time risk assessment result of reflux aspiration based on the preoperative information and the monitoring data using a risk assessment model; When the risk assessment results and / or the monitoring data meet the early warning rules, a risk intervention suggestion window will automatically pop up on the main interface, push clinical intervention suggestions and receive doctor feedback. The risk log viewing module is configured to display a risk log sub-interface in response to a trigger command on the risk log viewing control, supporting the retrospective review of risk assessment records, intervention interaction information, and doctor feedback during the examination process, in order to support the continuous optimization of the risk assessment model.

Images