Multifunctional touch display screen for operating room doorway

Abstract

This invention discloses a multi-functional touchscreen display at the entrance of an operating room, comprising a display body, a processor, a multimodal data acquisition unit, an AI control module, and a cascaded anomaly detection and hierarchical protection module. The multimodal data acquisition unit collects environmental data, equipment operation data, and image data of the display content. The AI ​​control module incorporates a mechanism-data hybrid model and an adaptive weighting unit, dynamically generating environmental control commands by fusing physical prediction residuals and deep learning probability entropy values. The cascaded anomaly detection and hierarchical protection module executes a three-level response strategy, achieving non-intrusive prompts, parameter correction, and equipment redundancy switching. Furthermore, it includes independent strip screens forming a dual-link redundant display to ensure continuous output of core vital signs parameters when the main link is interrupted. This invention achieves multimodal information fusion perception, intelligent control, and hierarchical anomaly warning, improving surgical safety and system reliability.

Description

Technical Field

[0001] This invention relates to the field of intelligent medical equipment technology, specifically to a multi-functional touch display screen at the entrance of an operating room. Background Technology

[0002] Currently, operating room entrances are typically equipped with touchscreen displays for showing surgical status, patient information, and environmental parameters. Their basic functions focus on information display and simple status prompts. With the continuous improvement of medical informatization and intelligence, modern operating rooms have gradually integrated various environmental monitoring sensors, vital sign monitoring equipment, and cleanroom air conditioning control systems. However, existing display systems are essentially still isolated information output terminals, lacking deep integration and interaction with equipment operating data, dynamic environmental changes, and medical imaging content within the operating room. This makes it difficult to achieve proactive perception and intelligent control of complex operating conditions within the operating room.

[0003] In the field of operating room environmental control and safety early warning, existing technical solutions mainly suffer from the following shortcomings: On the one hand, traditional environmental control logic is mostly based on single sensor threshold triggers, such as activating air conditioning after temperature or humidity exceeds limits. This passive response feedback regulation has significant response lag and cannot comprehensively consider the coupling interference of physical factors such as the thermal radiation of the operating lamp, personnel flow, and laminar airflow organization on sensor readings, resulting in insufficient control accuracy and high energy consumption. On the other hand, the identification of sudden abnormal events during surgery, such as electrosurgical smoke and insufflation machine leaks, generally relies on the subjective observation of medical staff or single-point monitoring alarms, lacking multimodal data fusion analysis and cascade early warning mechanisms, which can easily lead to missed or false alarms, posing potential risks to patient safety and surgical progress. In addition, existing display systems also have shortcomings in the redundant display of critical vital signs information. Once the main display link fails, core monitoring data will face the risk of interruption; the system architecture is rigid and lacks incremental learning capabilities, and surgical scheduling information is only presented in text form without participating in intelligent decision-making; multimodal data is not fully utilized and cannot support surgical scene recognition and intraoperative risk prediction.

[0004] Therefore, the present invention proposes a multi-functional touch display screen at the entrance of the operating room to at least partially solve the above-mentioned problems. Summary of the Invention

[0005] To address the aforementioned problems in the existing technology, this invention provides a multi-functional touch display screen at the operating room entrance, which solves the problem that existing operating room entrance touch display screens cannot actively sense and intelligently control complex working conditions inside the operating room.

[0006] The objective of this invention can be achieved through the following technical solutions: A multi-functional touch screen for operating room entrances, comprising a display body and a processor electrically connected to the display body, characterized in that it further comprises: A multimodal data acquisition unit is used to acquire environmental data, equipment operation data, and display screen content image data within the operating room. The AI ​​control module, running on the processor, is used to receive and process the multimodal data. The AI ​​control module has a built-in mechanism-data hybrid model and an adaptive weighting unit. The mechanism-data hybrid model generates environmental control instructions based on the multimodal data. A cascaded anomaly detection and hierarchical protection module, running on the processor, is used to identify abnormal events and execute preset hierarchical protection strategies based on the multimodal data and the output of the AI ​​control module.

[0007] As a preferred embodiment of the present invention, the mechanism-data hybrid model includes a mechanism kernel model and a data kernel network; The mechanistic kernel model is based on prior knowledge of the physical environment of the operating room to model and predict environmental data; The data kernel network uses deep learning methods to extract features and mine relationships from the multimodal data. The adaptive weighting unit is used to dynamically calculate the fusion weight of the prediction residual output by the mechanism kernel model and the probability entropy value output by the data kernel network after normalization, and generate the final environmental control command based on the weight.

[0008] As a preferred technical solution of the present invention, the mechanism core model is a mathematical model constructed based on computational fluid dynamics and thermodynamic laws. This model corrects and compensates for the airflow organization morphology of the operating room laminar flow purification system and the influence of the thermal radiation of the shadowless lamp on the temperature and humidity sensor readings. The prediction residual output by the mechanism kernel model is defined as the absolute value or root mean square error of the difference between the sensor measured value and the model predicted value.

[0009] As a preferred embodiment of the present invention, the data kernel network includes a computer vision model and a Transformer network; The computer vision model is used to extract vital sign waveforms and values ​​from the image data of the display screen content, and output the corresponding one-dimensional feature vector. The environmental data and the equipment operation data are normalized and then aligned with the one-dimensional feature vector in the time dimension and spliced ​​together to form a multimodal temporal feature matrix with a dimension of T×D. The Transformer network contains a multi-layer self-attention encoding structure, which models the temporal dependency relationship of the multimodal temporal feature matrix and maps it to the predicted values ​​of environmental parameters and / or the probability of device anomalies within a preset future time period through a fully connected output layer.

[0010] As a preferred embodiment of the present invention, the multimodal data acquisition unit includes a PM2.5 sensor; The AI ​​control module and / or the cascaded anomaly detection and hierarchical protection module are configured as follows: When the instantaneous rise slope of PM2.5 sensor data exceeds the first threshold and continues to exceed the second threshold, medical smoke events generated during electrosurgical procedures are identified. Based on the real-time concentration value of CO2 sensor data exceeding the third threshold and the duration exceeding the fourth threshold, combined with the operating status signal of the insufflator, high-pressure gas leakage events of the insufflator during laparoscopic surgery can be identified.

[0011] As a preferred embodiment of the present invention, the cascaded anomaly detection and hierarchical protection module is configured to execute a three-level response strategy based on multi-cause-one-effect logic, including: Level 1 Response: When the data from a single sensor deviates from the normal range but does not reach the device adjustment trigger threshold, and the confidence level of the fusion of the mechanism kernel prediction residual and the data kernel probability entropy value output by the AI ​​control module is lower than the first confidence threshold, a non-intrusive visual prompt dynamically associated with the degree of deviation is generated on the main body of the display screen. Level 2 response: When multiple sensor data change in correlation, and the AI ​​control module analyzes the data and points to a specific device malfunction or environmental anomaly, and the fusion confidence level is higher than the second confidence level threshold, parameter correction or device adjustment commands are automatically executed. Level 3 Response: When a serious event that may endanger patient safety or surgical procedure is detected, and the fusion confidence level is higher than the third confidence level threshold, the highest level of audible and visual alarm is triggered, and safety redundancy switching or shutdown protection of critical equipment is performed. The first confidence threshold, the second confidence threshold, and the third confidence threshold increase sequentially.

[0012] As a preferred embodiment of the present invention, the AI ​​control module further includes an incremental learning unit, which is configured as follows: By analyzing data from the hospital's surgical scheduling system and real-time equipment operation data, the current stage of the surgery can be determined, including anesthesia induction, surgical incision, or suturing closure. Based on the determination result, the data samples collected in the stage are assigned higher learning weights than regular samples, and the data kernel network is trained with a higher update frequency than usual.

[0013] As a preferred embodiment of the present invention, it further includes a redundant display system, the redundant display system comprising: An independent bar screen, which has a power supply module, an independent main control unit, and an independent video frame buffer that are independent of the main display screen body; The independent bar screen and the processor are connected in a dual-link redundant manner through a first communication link and a second communication link, and are equipped with a heartbeat monitoring program. When the heartbeat signal sent by the processor through the first communication link is interrupted for more than a preset threshold, the independent bar screen automatically switches the video signal source to its frame buffer and continuously displays at least one core vital sign parameter, such as heart rate and blood oxygen saturation, which are directly obtained from the medical device data interface, until the first communication link is restored.

[0014] As a preferred embodiment of the present invention, the computer vision model is a target detection model, used to locate and identify the display area from the content image data of the display screen; The multimodal data acquisition unit directly acquires heart rate, blood oxygen saturation, and waveform data from the monitor through the medical device data interface, and / or acquires images of the monitor display interface through the image acquisition device. The computer vision model then identifies the display area in the image and parses it into structured numerical and waveform data.

[0015] As a preferred embodiment of the present invention, the multimodal data further includes surgical scheduling data obtained from the hospital surgical scheduling information system, wherein the surgical scheduling data includes surgical type, attending physician and patient basic information; The data core network uses the surgical scheduling data as contextual features and embeds it together with the environmental data, equipment operation data, and display screen content image data to assist the model in surgical scene recognition and intraoperative risk prediction.

[0016] The beneficial effects of this invention are as follows: By fusing environmental, equipment, and image data through a multimodal data acquisition unit, and combining a mechanism-data hybrid model with an adaptive weighting unit, intelligent closed-loop control of the operating room environment is achieved, significantly improving the control accuracy and response speed of environmental parameters and effectively reducing energy consumption; through a cascaded anomaly detection and graded protection module, a three-level response strategy is executed based on fused confidence, providing non-invasive visual cues for minor anomalies, automatically correcting parameters for complex anomalies, and triggering audible and visual alarms and redundancy switching for serious events, greatly enhancing the safety of the surgical procedure; by using PM2.5 instantaneous slope and C O2 concentration combined with the status of the insufflator identifies electrosurgical smoke and insufflation leaks, filling the gap in automatic identification of specific intraoperative risks and ensuring the occupational exposure safety of medical staff. The incremental learning unit prioritizes training the data kernel network according to the surgical stage, enabling the system to adaptively optimize and continuously improve prediction accuracy. The redundant display system employs independent power supply, dual-link heart rate monitoring, and an autonomous switching mechanism to ensure continuous display of core vital signs even in the event of a main screen failure, significantly improving system reliability. Surgical scheduling data is embedded as contextual features to assist in surgical scenario identification and risk prediction, achieving a higher level of intelligent decision-making. In summary, this invention achieves a comprehensive intelligent upgrade of operating room environment control, anomaly protection, and information display, possessing extremely high clinical application value and promising prospects for widespread adoption. Attached Figure Description

[0017] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings.

[0018] Figure 1 This is a schematic diagram of the overall system architecture of the present invention.

[0019] Figure 2 This is a schematic diagram of the data core network processing flow of the present invention.

[0020] Figure 3 This is a schematic diagram of the three-level response decision-making process of the present invention. Detailed Implementation

[0021] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided.

[0022] Please see Figures 1 to 3This embodiment provides a multi-functional touch display screen for the operating room entrance. The display screen includes a main body and a processor electrically connected to the main body. The main body, serving as the primary human-machine interface, is installed at the operating room entrance or on the central control panel to display various information and receive touch operations. The processor, as the core computing unit, is embedded within the main body and is responsible for running various algorithm modules and controlling the collaborative operation of various hardware components.

[0023] The core improvement of the invention lies in the addition of a multimodal data acquisition unit, an AI control module, and a cascaded anomaly detection and hierarchical protection module.

[0024] The multimodal data acquisition unit is used to collect three types of core data within the operating room. The first type is environmental data, including temperature, humidity, pressure difference, PM2.5 concentration, and CO2 concentration. This data is acquired in real time through various sensors installed at the operating room's air supply and return vents and on the walls. The second type is equipment operation data, including the start / stop status and pressure parameters of the insufflator, the operating mode and setpoints of the air conditioning system, and the on / off status and brightness adjustment information of the operating lights. This data is acquired through communication with controlled devices within the operating room via RS485 bus or TCP / IP protocol. The third type is display screen image data, primarily obtained through two methods: first, by directly acquiring structured values ​​of heart rate, blood oxygen saturation, and waveform data from the monitor via the medical device data interface; second, by capturing screen images from the monitor's display interface using a camera to obtain visual images containing vital signs information.

[0025] The AI ​​control module runs on the processor and receives and processes all data collected by the multimodal data acquisition unit. This module incorporates a mechanism-data hybrid model and an adaptive weighting unit. The mechanism-data hybrid model can generate environmental control commands based on multimodal data, such as adjusting the air conditioner's set temperature, adjusting the brightness of the shadowless lamp, and controlling the pressure of the insufflator. The core of this model lies in the organic integration of mechanism modeling based on prior physical knowledge with deep learning based on big data, thereby achieving a more accurate and robust control effect than a single model.

[0026] The cascaded anomaly detection and graded protection module also runs on the processor. Based on multimodal data and the output of the AI ​​control module, it identifies various abnormal events that may occur in the operating room in real time and executes preset graded protection strategies according to the severity of the event. The module's input includes both raw multimodal sensor data and intermediate results calculated by the AI ​​control module. Through multi-dimensional correlation analysis, it achieves accurate identification of abnormal events and appropriate graded responses.

[0027] As a further improvement of this invention, the mechanism-data hybrid model specifically comprises two core components: a mechanism kernel model and a data kernel network. The mechanism kernel model models and predicts environmental data based on prior knowledge of the operating room's physical environment. This model utilizes computational fluid dynamics and thermodynamics laws to construct mathematical equations, simulating the physical processes of airflow and heat transfer within the operating room, thereby calculating the theoretical values ​​of each environmental parameter under ideal physical conditions. The data kernel network, on the other hand, uses deep learning methods to extract features and mine relationships from multimodal data. This network can learn complex nonlinear relationships between environmental parameters, equipment states, and the environment and equipment from massive amounts of historical data, thus discovering hidden patterns that are difficult to capture using only physical models. The adaptive weighting unit's role is to dynamically calculate the fusion weight of the predicted residuals output by the mechanism kernel model and the probability entropy values ​​output by the data kernel network after normalization, and then generate the final environmental control instructions based on these weights. When the prediction residual of the mechanistic kernel model is small, it indicates that the current operating condition matches the assumptions of the physical model well, and the weight of the mechanistic kernel model will be increased accordingly. Conversely, when the probability entropy value of the data kernel network is low, it indicates that the deep learning model's judgment of the current state is relatively certain, and the weight of the data kernel network will be increased accordingly. This dynamic weighting mechanism ensures that under any operating condition, the final control command will prioritize the model output with higher confidence.

[0028] More specifically, the mechanistic kernel model is a mathematical model built upon the laws of computational fluid dynamics and thermodynamics. This model specifically models the airflow organization morphology of the laminar flow purification system in the operating room, considering the influence of the airflow velocity and direction at the laminar flow supply vents and the location and distribution of the return air vents on the indoor temperature and humidity distribution. Simultaneously, the model also corrects and compensates for the impact of the thermal radiation from the operating lamp on the temperature and humidity sensor readings. The operating lamp generates significant thermal radiation during operation; if a temperature and humidity sensor is directly installed near the lamp, its readings will deviate due to radiative heating. The mechanistic kernel model calculates parameters such as the radiation angle coefficient and heat flux density between the operating lamp and the sensor, subtracting the additional temperature rise caused by thermal radiation from the sensor's measured values, thereby restoring the true ambient temperature. The prediction residual output by the mechanistic kernel model is defined as the absolute value or root mean square error of the difference between the sensor's measured value and the model's predicted value; this residual directly reflects the confidence level of the physical model under the current operating conditions.

[0029] The data core network further includes a computer vision model and a Transformer network. The computer vision model extracts vital sign waveforms and values ​​from the display screen content image data and outputs corresponding one-dimensional feature vectors. Specifically, this computer vision model employs object detection models, such as YOLOv7 or Faster R-CNN, to locate and identify display areas from the display screen content image data. The multimodal data acquisition unit directly acquires heart rate, blood oxygen saturation, and waveform data from the monitor via the medical device data interface, a preferred and highly reliable acquisition method. Simultaneously, as a supplementary or backup solution, the multimodal data acquisition unit also acquires images of the monitor's display interface via an image acquisition device, and the computer vision model identifies the display areas in the images and parses them into structured numerical and waveform data. When the monitor itself lacks a data output interface or the interface malfunctions, this camera recognition solution ensures continuous acquisition of vital sign data.

[0030] The one-dimensional feature vector output by the computer vision model represents vital sign information extracted from the image, such as a heart rate of 72 beats / min, blood oxygen saturation of 98%, and waveform sampling point sequence. Environmental data and equipment operation data are normalized and then concatenated with this one-dimensional feature vector along the time dimension. Since the sampling frequencies of the three types of data may differ, they are first aligned along the time axis through interpolation or resampling, then normalized separately to eliminate the influence of dimensions, and finally concatenated along the feature dimension to form a multimodal temporal feature matrix of dimension T×D. Here, T represents the time step, for example, a 60-second time window; D represents the feature dimension, i.e., the length of the concatenated feature vector at each time step. This matrix comprehensively depicts the synchronous evolution of the operating room environment, equipment status, and patient vital signs within a past time window.

[0031] The Transformer network receives this multimodal temporal feature matrix as input. The Transformer network contains a multi-layer self-attention encoding structure, enabling it to model the temporal dependencies of the input matrix. The self-attention mechanism allows the network to simultaneously focus on the correlations between different time steps, capturing long-distance temporal dependencies, such as the causal relationship between the increase in CO2 concentration in the previous 30 seconds and the change in insufflation machine pressure at the current moment. After multi-layer self-attention encoding, the Transformer network encodes the entire temporal matrix into a latent vector containing global information, which is then mapped through a fully connected output layer to predicted environmental parameters and / or equipment anomaly probabilities for a preset future time period. Predicted environmental parameters include, for example, temperature and humidity trends over the next 5 minutes; equipment anomaly probabilities include, for example, the probability of a leak in the insufflation machine within the next 10 minutes and the probability of a malfunction in the air conditioning system within the next 30 minutes. This predictive information provides crucial information for the feedforward control of the AI ​​regulation module and the early warning of the cascaded anomaly detection module.

[0032] Designed specifically to address surgical-specific risk events, the multimodal data acquisition unit includes a PM2.5 sensor for real-time monitoring of particulate matter concentration in the operating room. An AI control module or a cascaded anomaly detection and graded protection module is configured to identify medical smoke events generated during electrosurgical procedures when the instantaneous rise rate of PM2.5 sensor data exceeds a first threshold and continues to exceed a second threshold. Electrosurgical equipment generates a large amount of smoke containing fine particulate matter during operation. This smoke not only affects the surgical field of view but also contains harmful substances. By monitoring sudden changes in PM2.5 concentration, the system can detect smoke generation immediately and trigger corresponding ventilation and smoke extraction measures. Simultaneously, the system is also configured to identify high-pressure gas leakage events in the insufflator during laparoscopic surgery when the real-time concentration value of CO2 sensor data exceeds a third threshold and lasts for a duration exceeding a fourth threshold, combined with the insufflator's operating status signal. If a leak occurs in the tubing or a seal fails when the insufflator injects CO2 gas into the abdominal cavity, it can lead to an abnormally high CO2 concentration in the operating room. By jointly analyzing changes in CO2 concentration and the operating status of the pneumoperitoneum machine, the system can accurately distinguish pneumoperitoneum leakage from other CO2 sources and avoid false alarms.

[0033] The core innovation of the cascaded anomaly detection and hierarchical protection module lies in implementing a three-level response strategy based on multi-cause-one-effect logic. This strategy includes three progressively advancing protection levels: Level 1 response, Level 2 response, and Level 3 response. The first confidence threshold, the second confidence threshold, and the third confidence threshold increase sequentially, for example, set to 0.3, 0.6, and 0.8 respectively.

[0034] A Level 1 response is triggered when the following conditions are met simultaneously: a single sensor data deviation from the normal range has not yet reached the trigger threshold requiring device adjustment, and the fusion confidence level of the mechanism kernel prediction residual and the data kernel probability entropy value output by the AI ​​control module is lower than the first confidence threshold. This state indicates a slight anomaly in the sensor reading, and the AI ​​model's judgment of this anomaly has low confidence, potentially indicating sensor drift, transient interference, or model mismatch. In this case, the system does not display a large alarm window on the main interface, nor does it emit any beeping sounds. Instead, it generates a non-intrusive visual prompt dynamically correlated with the degree of deviation in the status bar area of ​​the main display screen. For example, when the deviation of the measured temperature value from the historical baseline average is 5% to 10%, a pale yellow breathing light icon is displayed in the status bar with a breathing cycle of 2 seconds; when the deviation is 10% to 15%, the icon switches to an orange pulse flashing with a flashing frequency of 1Hz; when the deviation exceeds 15% but has not yet reached the device adjustment trigger threshold, the icon switches to a red slow flashing with a flashing frequency of 0.5Hz. These visual cues remain active without interrupting the normal use of core operating interfaces such as surgical timer and vital signs display. They both convey warning information to medical staff and avoid interfering with the surgical process.

[0035] A Level 2 response is triggered when the following conditions are met simultaneously: multiple sensor data show correlated changes, and analysis by the AI ​​control module points to a specific equipment malfunction or environmental anomaly, while the fusion confidence level is higher than the second confidence threshold. At this point, the system's judgment of the abnormal event has a high confidence level, confirming an actual problem requiring intervention. The system no longer merely provides prompts but automatically executes parameter corrections or equipment adjustment commands. For example, when temperature sensors in multiple areas show temperature deviations and the AI ​​model analysis indicates insufficient cooling capacity of the air conditioning system, the system automatically sends a command to the air conditioning controller to lower the set temperature by 1 to 2 degrees Celsius. Similarly, when the pressure of the insufflator fluctuates and the CO2 concentration is synchronously abnormal, the system automatically sends a command to the insufflator to temporarily reduce the inflation rate to stabilize the pressure.

[0036] A Level 3 response is triggered when the following conditions are met simultaneously: the system detects a serious event that may endanger patient safety or the surgical procedure, and the fusion confidence level is higher than the third confidence threshold. At this time, the highest level audible and visual alarm is triggered, with a full-screen red flashing display accompanied by a continuous buzzing sound to ensure the attention of all medical staff in the operating room. Simultaneously, the system performs safety redundancy switching or shutdown protection for critical equipment. For example, when an anomaly is detected in the main power supply line and the backup power supply has not yet automatically switched on, the system immediately cuts off power to non-critical equipment to ensure power supply to life support equipment; when a serious leak is detected in the insufflator that cannot be automatically repaired, the system directly executes an emergency shutdown of the insufflator and automatically switches to the backup insufflator; when an impending failure is detected in the main display screen, the system proactively pushes key vital signs parameters to the redundant display system to ensure uninterrupted critical information.

[0037] As a further refinement of the above-mentioned three-level response strategy, the fusion confidence C in this embodiment is calculated as follows: C = α·(1-ê) + (1-α)·(1-H), where ê is the normalized mechanism kernel prediction residual, with a value range of [0, 1]; H is the normalized data kernel probability entropy value, with a value range of [0, 1]; and α is the fusion weight output by the adaptive weighting unit, dynamically determined by the current ê and H values. This calculation formula weights and harmonics the confidence of the physical model and the confidence of the data model to form a unified fusion confidence index, providing a quantitative basis for three-level response decision-making.

[0038] It is important to note that in this embodiment, the first confidence threshold θ1, the second confidence threshold θ2, and the third confidence threshold θ3 satisfy an increasing relationship of θ1 < θ2 < θ3. This setting reflects a counterintuitive but highly intelligent decision-making logic: low confidence triggers a prompt, and high confidence triggers intervention. In traditional anomaly response systems, a "deterministic incrementing" logic is usually followed, meaning that action is only taken when the system has a high degree of confidence in an event, and no response is given to uncertain events. This invention is the opposite. When the fusion confidence is lower than the first threshold (i.e., the system is most uncertain about the current state), it means that an unknown anomaly that the model has not covered may have occurred. At this time, a non-invasive visual prompt is triggered at the first level to draw the attention of medical staff. When the fusion confidence is higher than the second threshold, it indicates that the system has a high degree of confidence in the analysis of changes in the multi-sensor correlation. At this time, an automatic adjustment at the second level is triggered to actively intervene and prevent the anomaly from escalating. When the fusion confidence is higher than the third threshold, it indicates that the system has an extremely high degree of confidence in the judgment of a serious event. At this time, an emergency protection at the third level is triggered. This counterintuitive logic of "low-confidence prompts and high-confidence interventions" incorporates uncertainty itself into the intelligent decision-making system, achieving a qualitative change from passive response to proactive early warning.

[0039] The AI ​​control module further includes an incremental learning unit, enabling the system to continuously evolve. This unit analyzes data from the hospital's surgical scheduling system and real-time equipment operation data to first determine the current stage of the surgery, such as anesthesia induction, surgical incision, or suturing closure. The surgical scheduling system provides the surgery type, estimated duration, basic patient information, and real-time equipment operation data, such as the status of the anesthesia machine, the activation signal of the electrosurgical unit, and the count of suture needles, all used for stage determination. Based on this determination, the incremental learning unit assigns higher learning weights to data samples collected during critical surgical stages than to regular samples and prioritizes training the data kernel network at a higher update frequency. For example, during the surgical incision stage, when electrosurgical smoke is generated more frequently, the system increases the sampling weight of PM2.5-related data for this stage and updates the network layer parameters related to smoke recognition in the data kernel network at twice the regular frequency, thereby continuously optimizing the model's predictive ability for this high-risk stage.

[0040] A redundant display system is another important safety feature of this invention. This redundant display system includes an independent bar screen, which has a power supply module, a separate main control unit, and a separate video frame buffer, independent of the main display screen. The independent bar screen and the processor are connected in a dual-link redundant manner via a first communication link and a second communication link. The first communication link can use TCP / IP Ethernet for transmitting regular synchronous display information; the second communication link can use an RS485 bus as a backup link for transmitting heartbeat signals and emergency data. The system is equipped with a heartbeat monitoring program. The main display screen periodically sends heartbeat signals to the independent bar screen via the first communication link, for example, once per second. When the independent bar screen detects that the heartbeat signal interruption of the first communication link exceeds a preset threshold, for example, if no heartbeat packet is received for 3 consecutive seconds, it determines that the main display screen has malfunctioned or the first communication link has been interrupted. At this time, the independent bar screen autonomously switches the video signal source to its internal video frame buffer, no longer relying on the main screen to push data, but directly obtaining at least one core vital sign parameter, such as heart rate and blood oxygen saturation, from the medical device data interface, and continuously displaying this critical information on the bar screen. This autonomous switching process requires no manual intervention and takes only milliseconds, ensuring the continuous readability of core monitoring data in the event of a main screen failure. After switching, the bar screen will continue to display vital sign parameters until the first communication link is restored and the main display screen returns to normal operation.

[0041] As a further deepening of the utilization of multimodal data, multimodal data also includes surgical scheduling data obtained from the hospital's surgical scheduling information system. This data includes surgical types such as cardiac surgery, neurosurgery, and laparoscopic surgery, surgeon information, and basic patient information such as age, weight, and past medical history. The data kernel network uses this surgical scheduling data as contextual features and jointly embeds it with environmental data, equipment operation data, and display screen content image data. Joint embedding refers to mapping different types of features to the same vector space for fusion, enabling the model to understand the specific environmental parameter requirements of different surgical types. For example, cardiac surgery has extremely high requirements for temperature stability, neurosurgery has strict requirements for cleanliness, and pediatric surgery has specific restrictions on light intensity. By using surgical types as contextual features, the data kernel network can adjust the prediction model and behavioral strategies for different surgical scenarios, assisting the system in more accurate surgical scenario identification and more reliable intraoperative risk prediction.

[0042] In summary, this invention achieves comprehensive perception of environmental, equipment, and image data through a multimodal data acquisition unit; organically integrates and dynamically weights mechanistic and deep learning models through an AI control module; implements three levels of intelligent protection—from non-intrusion alerts to redundancy switching—through a cascaded anomaly detection module; enables adaptive optimization for the surgical stage through an incremental learning unit; ensures continuous protection of critical information in the event of a main screen failure through a redundant display system; and achieves surgical scenario perception and risk prediction through contextual embedding of surgical scheduling data. These technical solutions collectively address the technical problem that existing touchscreen displays at operating room entrances cannot proactively perceive and intelligently control complex operating conditions, achieving a comprehensive intelligent upgrade of operating room environmental control, anomaly protection, and information display.

[0043] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A multi-functional touch display screen for operating room entrances, comprising a display screen body and a processor electrically connected to the display screen body, characterized in that, Also includes: A multimodal data acquisition unit is used to acquire environmental data, equipment operation data, and display screen content image data within the operating room. The AI ​​control module, running on the processor, is used to receive and process the multimodal data. The AI ​​control module has a built-in mechanism-data hybrid model and an adaptive weighting unit. The mechanism-data hybrid model generates environmental control instructions based on the multimodal data. A cascaded anomaly detection and hierarchical protection module, running on the processor, is used to identify abnormal events and execute preset hierarchical protection strategies based on the multimodal data and the output of the AI ​​control module.

2. The multi-functional touch screen display at the operating room entrance according to claim 1, characterized in that, The mechanism-data hybrid model includes a mechanism kernel model and a data kernel network; The mechanistic kernel model is based on prior knowledge of the physical environment of the operating room to model and predict environmental data; The data kernel network uses deep learning methods to extract features and mine relationships from the multimodal data. The adaptive weighting unit is used to dynamically calculate the fusion weight of the prediction residual output by the mechanism kernel model and the probability entropy value output by the data kernel network after normalization, and generate the final environmental control command based on the weight.

3. The multi-functional touch screen display at the operating room entrance according to claim 2, characterized in that, The mechanism core model is a mathematical model built on computational fluid dynamics and thermodynamic laws. This model corrects and compensates for the airflow organization morphology of the operating room laminar flow purification system and the influence of the thermal radiation of the shadowless lamp on the temperature and humidity sensor readings. The prediction residual output by the mechanism kernel model is defined as the absolute value or root mean square error of the difference between the sensor measured value and the model predicted value.

4. The multi-functional touch screen display at the operating room entrance according to claim 2, characterized in that, The data core network includes a computer vision model and a Transformer network; The computer vision model is used to extract vital sign waveforms and values ​​from the image data of the display screen content, and output the corresponding one-dimensional feature vector. The environmental data and the equipment operation data are normalized and then aligned with the one-dimensional feature vector in the time dimension and spliced ​​together to form a multimodal temporal feature matrix with a dimension of T×D. The Transformer network contains a multi-layer self-attention encoding structure, which models the temporal dependency relationship of the multimodal temporal feature matrix and maps it to the predicted values ​​of environmental parameters and / or the probability of device anomalies within a preset future time period through a fully connected output layer.

5. The multi-functional touch screen display at the operating room entrance according to claim 1, characterized in that, The multimodal data acquisition unit includes a PM2.5 sensor; The AI ​​control module and / or the cascaded anomaly detection and hierarchical protection module are configured as follows: When the instantaneous rise slope of PM2.5 sensor data exceeds the first threshold and continues to exceed the second threshold, medical smoke events generated during electrosurgical procedures are identified. Based on the real-time concentration value of CO2 sensor data exceeding the third threshold and the duration exceeding the fourth threshold, combined with the operating status signal of the insufflator, high-pressure gas leakage events of the insufflator during laparoscopic surgery can be identified.

6. The multi-functional touch screen display at the operating room entrance according to claim 1, characterized in that, The cascaded anomaly detection and hierarchical protection module is configured to execute a three-level response strategy based on multi-cause-one-effect logic, including: Level 1 Response: When the data from a single sensor deviates from the normal range but does not reach the device adjustment trigger threshold, and the confidence level of the fusion of the mechanism kernel prediction residual and the data kernel probability entropy value output by the AI ​​control module is lower than the first confidence threshold, a non-intrusive visual prompt dynamically associated with the degree of deviation is generated on the main body of the display screen. Level 2 response: When multiple sensor data change in correlation, and the AI ​​control module analyzes the data and points to a specific device malfunction or environmental anomaly, and the fusion confidence level is higher than the second confidence level threshold, parameter correction or device adjustment commands are automatically executed. Level 3 Response: When a serious event that may endanger patient safety or surgical procedure is detected, and the fusion confidence level is higher than the third confidence level threshold, the highest level of audible and visual alarm is triggered, and safety redundancy switching or shutdown protection of critical equipment is performed. The first confidence threshold, the second confidence threshold, and the third confidence threshold increase sequentially.

7. The multi-functional touch screen display at the operating room entrance according to claim 4, characterized in that, The AI ​​control module further includes an incremental learning unit, which is configured as follows: By analyzing data from the hospital's surgical scheduling system and real-time equipment operation data, the current stage of the surgery can be determined, including anesthesia induction, surgical incision, or suturing closure. Based on the determination result, the data samples collected in the stage are assigned higher learning weights than regular samples, and the data kernel network is trained with a higher update frequency than usual.

8. The multi-functional touch screen display at the operating room entrance according to claim 1, characterized in that, It also includes a redundant display system, which comprises: An independent bar screen, which has a power supply module, an independent main control unit, and an independent video frame buffer that are independent of the main display screen body; The independent bar screen and the processor are connected in a dual-link redundant manner through a first communication link and a second communication link, and are equipped with a heartbeat monitoring program. When the heartbeat signal sent by the processor through the first communication link is interrupted for more than a preset threshold, the independent bar screen automatically switches the video signal source to its frame buffer and continuously displays at least one core vital sign parameter, such as heart rate and blood oxygen saturation, which are directly obtained from the medical device data interface, until the first communication link is restored.

9. The multi-functional touch screen display at the operating room entrance according to claim 4, characterized in that, The computer vision model is a target detection model, used to locate and identify the display area from the content image data of the display screen; The multimodal data acquisition unit directly acquires heart rate, blood oxygen saturation, and waveform data from the monitor through the medical device data interface, and / or acquires images of the monitor display interface through the image acquisition device. The computer vision model then identifies the display area in the image and parses it into structured numerical and waveform data.

10. The multi-functional touch display screen at the operating room entrance according to claim 4, characterized in that, The multimodal data also includes surgical scheduling data obtained from the hospital's surgical scheduling information system, which includes surgical type, attending physician, and basic patient information. The data core network uses the surgical scheduling data as contextual features and embeds it together with the environmental data, equipment operation data, and display screen content image data to assist the model in surgical scene recognition and intraoperative risk prediction.

Images