A reflux identification device of an intelligent gastric tube

Abstract

This invention provides an intelligent reflux detection device for a gastric tube, comprising a gastric tube inserted into the patient's esophagus via a laryngeal mask airway, and a video information acquisition device mounted on the gastric tube. The video information acquisition device includes a camera and a sensor unit for acquiring video data within the patient's esophagus. A data processing module is used to analyze the video data frame-by-frame and detect reflux material, marking the location of the reflux material and generating reflux data. A suction module, connected to the data processing module, is used to select different suction strengths based on the reflux data. This intelligent reflux detection device for a gastric tube, through the configuration of a camera and sensor unit and the target detection and time series prediction model integrated into the data processing module, can accurately identify the location and magnitude of reflux material and predict its dynamic trend. Through the multi-level strength adjustment and graded alarm mechanism of the suction module, it can achieve intelligent suction of reflux material and risk warning.

Description

Technical Field

[0001] This utility model relates to the field of medical devices, and in particular to a reflux identification device for an intelligent gastric tube. Background Technology

[0002] Currently, with the development of medical technology, general anesthesia has become an important part of modern surgery. During anesthesia, doctors typically need to correctly insert a laryngeal mask airway into the patient's glottis and drain stomach contents through a gastric tube to reduce the risk of reflux into the trachea. However, due to differences in patient position, the type of stomach contents, and the precision of laryngeal mask airway insertion, the risk of stomach contents and gastric juices refluxing into the esophagus and even the trachea still exists. This reflux can lead to suffocation, aspiration pneumonia, or even death. Traditional solutions often use ordinary gastric tubes and negative pressure suction devices for drainage, but when the reflux volume is large or the reflux rate is too fast, manual intervention is insufficient to handle it promptly, posing significant safety hazards.

[0003] In recent years, visualization technology has been increasingly applied to the design of gastric tubes to improve the precision of their operation. By integrating miniature optical lenses inside the tube, doctors can observe the insertion process and position in real time, significantly improving the controllability of the procedure. However, these visualized gastric tubes have limited functionality; they can only assist doctors in observing the tube's position and cannot intelligently identify and handle reflux. Especially during surgery, when it is necessary to quickly determine the amount of reflux and apply appropriate suction, relying on manual experience for adjustment can easily lead to operational errors and increase the risk of patient injury.

[0004] Driven by artificial intelligence, deep learning-based object detection and time series prediction methods have shown great promise in the medical field. Algorithms such as the YOLO object detection model can quickly identify the spatial location of refluxed material, while convolutional LSTM models can predict the changing trends of refluxed material based on historical data, providing technical support for the intelligentization of medical devices. However, these algorithms are currently mostly used for offline analysis and have not been fully integrated with actual equipment, still lacking in real-time performance, automated operation, and adaptability to the surgical environment. Therefore, how to combine intelligent detection algorithms with visualization technology to develop an intelligent gastric tube device that can monitor, accurately identify, and automatically process refluxed material in real time has become an urgent problem to be solved in the current technological field. Utility Model Content

[0005] To address the aforementioned technical problems, this utility model provides an intelligent gastric tube reflux identification device.

[0006] This invention provides an intelligent reflux detection device for a gastric tube, comprising: a gastric tube inserted into a patient's esophagus via a laryngeal mask, the gastric tube being equipped with a video information acquisition device; the video information acquisition device including a camera and a sensor unit for acquiring video data within the patient's esophagus; a data processing module connected to the video information acquisition device for frame-by-frame analysis of the video data and detection of refluxed material, marking the location of the refluxed material and generating reflux data; a suction module connected to the data processing module for selecting different suction strengths based on the reflux data; and an alarm control module connected to the data processing module for setting alarm states based on the reflux data.

[0007] Furthermore, the data processing module specifically includes: a data reading unit, used to read the video data frame by frame; a target detection unit, used to input the frame-by-frame images into a pre-trained YOLO image detection model, identify the re-flowing material and mark its specific location; and a secondary analysis unit, used to perform secondary target extraction on the marked re-flowing material images, analyze the magnitude of the re-flowing material and generate re-flowing data.

[0008] Furthermore, the data processing module specifically includes: a multi-frame processing unit, used to select multiple consecutive frames of images and input them into a convolutional LSTM model to predict the state of the reef in future frames; a change analysis unit, used to analyze the changing trend and spatial distribution characteristics of the reef quantity based on the prediction results of the convolutional LSTM model; and a visualization unit, used to annotate the predicted bounding boxes of the reef and generate visualization results.

[0009] Furthermore, the data processing module is specifically capable of generating first return flow data, second return flow data, or third return flow data based on the size ratio of the returned material; the attraction module specifically includes: a force control unit, used to select a first-level attraction intensity based on the first return flow data, a second-level attraction intensity based on the second return flow data, and a third-level attraction intensity based on the third return flow data; the alarm control module specifically includes: an alarm level setting unit, used to set the alarm status to a slight alarm, a moderate alarm, or a severe alarm respectively based on the first return flow data, the second return flow data, and the third return flow data; and an alarm output unit, used to output alarm signals in the form of sound, light, or vibration.

[0010] This invention discloses an intelligent reflux identification device for gastric tubes. By incorporating a video information acquisition device, a data processing module, a suction module, and an alarm control module within the gastric tube, it achieves accurate identification, intelligent processing, and real-time alarm for refluxed material, significantly improving the safety and efficiency of intraoperative reflux treatment. Through the configuration of a high-resolution miniature optical lens and sensor unit, combined with the application of a transparent protective sheet, it can acquire real-time video data of the patient's esophagus, providing clear and stable imaging results. This design enables visualization of the gastric tube insertion process, significantly reducing the error rate of the insertion operation and providing reliable assurance for intraoperative monitoring.

[0011] The data processing module integrates target detection and time series prediction models, enabling it to accurately identify the spatial location and magnitude of refluxed material. Simultaneously, it predicts the dynamic trends of refluxed material through multi-frame image analysis. This function allows medical staff to anticipate the future state of refluxed material, avoiding the emergency risks caused by sudden large amounts of refluxed material, and further improving the safety and reliability of the surgery.

[0012] This invention also features three levels of suction intensity—mild, moderate, and severe—based on the amount of reflux, with automated adjustment via a suction module. Compared to traditional manual adjustment of suction intensity, this device precisely matches the suction intensity to the amount of reflux, preventing reflux residue due to insufficient suction and avoiding secondary damage caused by excessive suction, thus significantly improving the efficiency and safety of the suction process. Furthermore, the alarm control module provides three levels of alarm status—mild, moderate, and severe—based on the amount and trend of reflux, alerting medical staff via sound, light, vibration, and SMS to ensure timely response to intraoperative risks. This multi-level alarm mechanism effectively reduces the burden on medical staff and provides reliable support for the handling of reflux during surgery. The dual-lumen gastric tube structure integrates video acquisition and reflux drainage functions. Combined with a high-speed data transmission module and optimized algorithms, this ensures the system's real-time processing capability and stability, providing efficient and reliable technical support for reflux handling during surgery.

[0013] Other features and advantages of this invention 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 invention. The objects and other advantages of this invention may be realized and obtained by means of the structures particularly pointed out in the written description, claims, and drawings.

[0014] The technical solution of this utility model will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0015] The accompanying drawings are provided to further illustrate the present invention and form part of the specification. They are used together with the embodiments of the present invention to explain the present invention, but do not constitute a limitation thereof. In the drawings:

[0016] Figure 1 This is a schematic diagram of the first process of the reflux identification method for the intelligent gastric tube described in this utility model;

[0017] Figure 2 This is a schematic diagram of the second process of the reflux identification method for the intelligent gastric tube described in this utility model;

[0018] Figure 3 This is a schematic diagram of the third process of the reflux identification method for the intelligent gastric tube described in this utility model;

[0019] Figure 4 This is a first structural schematic diagram of the reflux identification device for the intelligent gastric tube described in this utility model;

[0020] Figure 5 This is a second structural schematic diagram of the reflux identification device for the intelligent gastric tube described in this utility model;

[0021] Figure 6 This is a schematic diagram of the third structure of the reflux identification device for the intelligent gastric tube described in this utility model;

[0022] Figure 7 This is a fourth structural schematic diagram of the reflux identification device for the intelligent gastric tube described in this utility model. Detailed Implementation

[0023] The structure and working principle of this utility model will be further explained below with reference to the accompanying drawings.

[0024] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings, but this is not intended to limit the present invention.

[0025] It should be understood that various modifications can be made to the embodiments disclosed herein. Therefore, the following description should not be considered as limiting, but merely as an example of embodiments. Other modifications within the scope and spirit of this disclosure will be apparent to those skilled in the art.

[0026] The accompanying drawings, which are included in and form part of this specification, illustrate embodiments of the present disclosure and, together with the general description of the disclosure given above and the detailed description of the embodiments given below, serve to explain the principles of the disclosure.

[0027] These and other features of the present invention will become apparent from the following description of preferred forms of embodiments given as non-limiting examples, with reference to the accompanying drawings.

[0028] It should also be understood that although the present invention has been described with reference to some specific examples, those skilled in the art can certainly implement many other equivalent forms of the present invention, which have the features described in the claims and are therefore all within the scope of protection defined herein.

[0029] The above and other aspects, features and advantages of this disclosure will become more apparent when taken in conjunction with the accompanying drawings and in view of the following detailed description.

[0030] Specific embodiments of the present disclosure are described thereafter with reference to the accompanying drawings; however, it should be understood that the disclosed embodiments are merely examples of the present disclosure and can be implemented in various ways. Well-known and / or repeated functions and structures are not described in detail to avoid unnecessary or redundant details that could obscure the present disclosure. Therefore, the specific structural and functional details disclosed herein are not intended to be limiting, but merely to serve as the basis and representative basis for the claims to teach those skilled in the art to use the present disclosure in a variety of substantially any suitable detailed structures.

[0031] This specification may use the phrases “in one embodiment,” “in another embodiment,” “in yet another embodiment,” or “in still another embodiment,” all of which may refer to one or more of the same or different embodiments according to this disclosure.

[0032] The embodiments of this utility model will now be described in detail with reference to the accompanying drawings.

[0033] like Figures 1 to 4 As shown, Figure 1 This is a schematic diagram of the first process of the reflux identification method for the intelligent gastric tube described in this utility model; Figure 2 This is a schematic diagram of the second process of the reflux identification method for the intelligent gastric tube described in this utility model; Figure 3 This is a schematic diagram of the third process of the reflux identification method for the intelligent gastric tube described in this utility model; Figure 4 This is a first structural schematic diagram of the intelligent gastric tube reflux identification device of the present invention; an intelligent gastric tube reflux identification method includes a gastric tube and a laryngeal mask, wherein a video information acquisition device is provided on the gastric tube, and the method includes the following steps:

[0034] S1 Inserts the gastric tube into the patient's esophagus through the laryngeal mask;

[0035] S2 acquires video data from inside the esophagus; in a preferred embodiment of this invention, the video information acquisition device specifically includes a miniature optical lens, a sensor unit, a data connection module, and supporting software. The miniature optical lens is the core component of the device, with a diameter less than 1.6mm. It uses a high-resolution CMOS chip as the imaging core, providing high-definition video acquisition capability of 1920×1080 pixels, while possessing a wide field of view of 120° to cover the entire esophagus. The lens employs an autofocus function, automatically adjusting the imaging focal length according to different distances inside the esophagus to ensure consistently clear and stable imaging. Furthermore, to adapt to the intraoperative environment, a transparent protective film is added to the front of the lens. The surface of the protective film has a hydrophobic coating, effectively preventing the adhesion of surgical secretions and avoiding impact on imaging results. The accompanying sensor unit receives the optical signals acquired by the optical lens in real time and performs brightness adjustment, noise suppression, and image optimization processing. The sensor has a built-in dynamic range extension (HDR) function, which significantly improves image brightness in low-light conditions during surgery. Simultaneously, it employs 3D noise reduction technology to filter each frame of the image, further enhancing image quality. This video information acquisition device transmits acquired image data to the data processing module in real time via a high-speed data connection module. The connection cable features a shielded design to prevent electromagnetic interference during surgery, achieving a transmission rate of up to 1Gbps and ensuring no data loss at a acquisition frequency of 30 frames per second. It also supports standardized interfaces (such as USB 3.0 or HDMI) for easy compatibility with various medical devices. During data transmission, the supported software system optimizes the image data in real time, including histogram equalization to enhance contrast and automatic white balance adjustment for color restoration. Furthermore, the video stream can be directly input into the target detection algorithm (YOLOv8) model to achieve real-time identification and location of regurgitated materials, which are then dynamically presented in high-definition on the monitoring terminal for easy observation and intervention by medical personnel. Experiments using a simulated human body model have verified that the video information acquisition device can provide high-resolution, low-noise, clear images in low-light environments, with a dynamic response time of less than 20ms and stable data transmission without frame loss, fully meeting the needs of precise intraoperative monitoring and data acquisition.

[0036] S3 reads the video data;

[0037] S4 analyzes the video data frame by frame and detects backflow.

[0038] S5 If a backflow object is found in a certain frame, mark the location of the backflow object;

[0039] S6 performs secondary target extraction on the frame, analyzes the magnitude of the reflow material, and generates reflow data.

[0040] S7 selects the corresponding attraction level and alarm status based on the backflow data;

[0041] In a preferred embodiment of this utility model, steps S4 and S5 further include the following sub-steps:

[0042] S31 inputs frame-by-frame images as test data into the established YOLO image detection model. In this embodiment, frame-by-frame images acquired by the video information acquisition device are transmitted to the data processing module through the data connection module and loaded as input data into the pre-trained YOLO image detection model. The YOLO model adopts an end-to-end detection framework, and its feature extraction part is composed of a multi-layer convolutional neural network (CNN), which can quickly and efficiently extract key features in video frames. The input image data needs to be standardized first to adjust the pixel value range to [0,1] and uniformly adjust the resolution to match the input dimension of the model (e.g., 416×416 pixels). This preprocessing step ensures the model's consistent processing capability for images from different sources, while significantly reducing the computational burden.

[0043] S32 utilizes the YOLO image detection model to analyze the input image, identify, and predict whether re-entry material exists in the image. In this embodiment, after inputting the image frame by frame, the YOLO model performs target detection analysis on each frame. Feature maps are extracted using a feature extraction network, and the model divides the image into multiple grid cells. Each cell predicts whether it contains a target (i.e., re-entry material), its location, and category. The detection model calculates the confidence score for each grid cell using the following formula: P(Object) × IoU = C

[0044] Where P(Object) is the probability of a backflow object existing within a grid cell, IoU is the Intersection over Union (IoU) between the predicted and ground truth bounding boxes, and C is the final confidence score. If the confidence score exceeds a set threshold (e.g., 0.5), a backflow object is considered detected. The model also predicts the category information of the backflow object to achieve accurate classification.

[0045] If S33 is correct, then the specific location of the returned object is determined; when the YOLO detection model confirms the presence of a returned object in the frame, the bounding box position of the returned object (top left corner coordinates (x, y), width w, and height h) is predicted by a regression algorithm. The specific calculation formula is as follows:

[0046] x=σ(t x )+c x y=σ(t) y )+c y ,

[0047] Where tx, ty, tw, th are the predicted values ​​output by the network, (cx, cy) are the coordinates of the top-left corner of the grid cell, (pw, ph) are the prior box size, and σ is the sigmoid activation function. The final generated prediction box can accurately mark the specific location of the return flow and provide its size information for subsequent processing.

[0048] S34 calculates the inference time of the image to evaluate the model's real-time performance and efficiency. In this embodiment, after completing object detection, the system records the inference time of the YOLO model, i.e., the total time T from loading the input image to outputting the prediction result. inf Inference time can be broken down into the following two parts:

[0049] T inf =T pre +T model

[0050] Where Tpre represents the image preprocessing time and Tmodel represents the model inference time. The model's processing efficiency in a real-time environment is evaluated by statistically analyzing the average inference time. Experimental results show that the average inference time of the YOLO model is 20ms, which can meet the high-frequency frame rate requirements (e.g., 30 frames / second) during surgery, achieving real-time target detection.

[0051] S35 generates and labels corresponding predicted bounding boxes on the image, visualizing the detection results. In this embodiment, after detecting reflux, the system overlays the predicted boxes onto the original image in a visual format, and uses different colors to label the type and magnitude of the reflux. For example, a green border is used for mild reflux, a yellow border for moderate reflux, and a red border for severe reflux. The bounding box positions are determined by the aforementioned predicted values ​​(x, y, w, h) and are dynamically updated on the monitoring terminal to create a real-time visualization effect. In addition, a prediction confidence score is overlaid in the visualization results and displayed at the top of each bounding box to help medical personnel quickly determine the reliability of the detection results.

[0052] In another preferred embodiment of this utility model, steps S4 and S5 further include the following sub-steps:

[0053] S31' selects 18 consecutive frames of images as input data. In this embodiment, to analyze the temporal variation trend of the refluxing material, the system selects 18 consecutive frames of images as input data from the frame sequence captured in real time by the video information acquisition device. These 18 frames correspond to approximately 1.5 hours of refluxing material status (assuming a frame interval of 5 minutes), which can fully reflect the variation characteristics of the refluxing material in different time periods. To ensure the quality of the input data, the system preprocesses each frame of image, including normalization to adjust the pixel values ​​to [0,1] and adjusting the image resolution to the size required by the model input (such as 128×128 or other suitable sizes). This preprocessing operation can reduce computational complexity and improve the prediction accuracy of the model.

[0054] S32' inputs the 18 frames of images into a pre-trained convolutional LSTM model, using the convolutional LSTM model to predict the state of the next 18 frames of images; in this embodiment, the preprocessed 18 frames of images are input into the convolutional LSTM (ConvLSTM) model to predict the state of the repatriated material in the next 18 frames of images. The convolutional LSTM model combines the characteristics of convolutional neural networks (CNN) and long short-term memory networks (LSTM), and can simultaneously capture the spatial and temporal features of the image. For each time step ttt, the state update formula of the ConvLSTM unit is as follows:

[0055] I t =σ(W i X t +U i H t-1 +b i )

[0056] F t =σ(W f X t +U f H t-1 +b f )

[0057] O t =σ(W o x t +U o H t-1 +b0)

[0058]

[0059] H t =O t ⊙tanh(C t )

[0060] Among them, \(x_t\) represents the current input image data, \(H_{t - 1}\) is the output state of the hidden layer at the previous moment, \(I_t\), \(F_t\), and \(O_t\) are the input gate, forget gate, and output gate respectively, \(*\) represents the convolution operation, \(W\) and \(U\) are weight matrices, \(b\) is the bias, \(\sigma\) is the Sigmoid activation function, and tanh is the hyperbolic tangent function. Through the above formula, the model can accurately predict future frames by combining spatial and temporal features.

[0061] S33’ returns the prediction results of the future 18-frame images, and analyzes the status of the reflux in the future 18 images, including the change trend of the reflux magnitude and its spatial distribution characteristics; the prediction results output by the convolutional LSTM model contain the reflux distribution and status information of the future 18-frame images. For each frame, the system analyzes the change trend of the reflux magnitude and the spatial distribution characteristics. The reflux magnitude is calculated by the ratio of the area of the predicted bounding box to the total area of the image.

[0062] S34’ annotates the corresponding predicted bounding box and visualizes the prediction results. For each frame of the predicted image, the system superimposes the predicted bounding box at the reflux position and annotates the magnitude value and the change trend information.

[0063] In a preferred embodiment of the present invention, in step S6, the frame is subjected to secondary target extraction, the reflux magnitude is analyzed, and reflux data is generated. In S7, the corresponding suction force and alarm status are selected according to the reflux data; specifically including:

[0064] The system selects appropriate suction force and alarm status through logical rules. The reflux magnitude LLL is divided into three intervals: mild (\(L\leq0.15\)), moderate (\(0.15\lt L\leq0.55\)), and severe (\(L\gt0.55\)). The specific logic is as follows:

[0065] When \(L\leq0.15\), the system generates the first reflux data, selects the first-level suction force to discharge the reflux in a gentle manner, and sets the alarm status to a mild alarm (green signal);

[0066] When \(0.15\lt L\leq0.55\), the system generates the second reflux data, selects the second-level suction force to discharge the reflux at a medium intensity, and sets the alarm status to a moderate alarm (yellow signal);

[0067] When \(L\gt0.55\), the system generates the third reflux data, selects the third-level suction force to quickly discharge the reflux at the maximum intensity, and sets the alarm status to a severe alarm (red signal).

[0068] The alarm mechanism synthesizes the magnitude data and the change rate information through the early warning decision function, specifically as follows:

[0069]

[0070] Where Bi is the reflux material status index, wi is the weighting coefficient, f(Bi) is the scoring function of the status index, θ is the threshold, and σ is the Sigmoid activation function. The alarm signal generated by this function is transmitted to medical staff through sound and light, vibration, or SMS to ensure timely response.

[0071] like Figures 5 to 7 As shown, Figure 5 This is a second structural schematic diagram of the reflux identification device for the intelligent gastric tube described in this utility model; Figure 6 This is a schematic diagram of the third structure of the intelligent gastric tube reflux identification device of this utility model. Figure 7 This is a fourth structural schematic diagram of the intelligent gastric tube reflux identification device of this utility model. The intelligent gastric tube reflux identification device of this utility model includes a gastric tube 2, a video information acquisition device 3, a data processing module, a suction module, an alarm control module, and a laryngeal mask 1. The modules and units are connected via signal or data transmission lines, working together to achieve accurate identification, suction, and real-time alarm functions for refluxed material. The gastric tube 2 is made of flexible medical plastic and has a double-lumen structure. The upper cavity accommodates the video information acquisition device and related sensor components, while the lower cavity serves as the discharge channel for refluxed material. A transparent protective sheet is provided at the front end of the gastric tube to protect the camera and sensors from interference from surgical secretions, while ensuring a clear imaging field. The rear end of the gastric tube 2 is connected to a negative pressure suction device via a gastric tube connector.

[0072] The video information acquisition device 3 is installed in the upper cavity of the gastric tube and includes a miniature optical lens and a sensor unit. The miniature optical lens uses a high-resolution CMOS chip, supporting high-definition imaging of 1920×1080 pixels and a field of view (FOV) of 120°, enabling comprehensive capture of the internal environment of the esophagus. The sensor unit acquires image data in real time and optimizes image brightness, contrast, and noise. This device is connected to the data processing module via a high-speed data connection module to ensure stable real-time transmission of video data.

[0073] The data processing module is the core unit of the intelligent gastric tube, used to analyze frame-by-frame image data transmitted by the video information acquisition device. The data processing module specifically includes:

[0074] Data reading unit: responsible for reading video data frame by frame and performing preprocessing operations such as normalization and resolution adjustment;

[0075] The object detection unit analyzes the image using the YOLO object detection model to identify and mark the location of the re-entrant material. The detection model determines the specific location and size of the re-entrant material by predicting bounding boxes, as shown in the following formula:

[0076] x=σ(t x )+c xy=σ(t) y )+c y ,

[0077] Where x,yx,yx,y are the center coordinates of the bounding box, and w,hw,hw,h are the width and height.

[0078] Multi-frame processing unit: Analyzes multiple consecutive frames of images and uses a deep learning model to predict the state of the return flow over a future period, including the trend of magnitude change and spatial distribution characteristics;

[0079] Secondary analysis unit: For the initially marked backflow area, extract more precise feature information, further calculate the magnitude and rate of change of the backflow, and thus generate backflow data describing the dynamic characteristics of the backflow.

[0080] The suction module is used to select the appropriate suction level based on the reflux data. Depending on the amount of reflux, the suction level is divided into three levels: light, medium, and heavy, to match the removal needs of different reflux volumes. Light suction is suitable for removing small amounts of reflux, medium suction is suitable for removing medium-sized reflux, and heavy suction is used to quickly process large-scale reflux, ensuring removal efficiency.

[0081] The alarm control module includes an alarm level setting unit and an alarm output unit. Based on the amount and dynamic trend of the refluxed material, this module sets mild, moderate, or severe alarm states and sends alarm signals to medical staff via sound, light, vibration, or SMS. A mild alarm typically alerts to a small amount of refluxed material, a moderate alarm warns of an increased reflux volume, and a severe alarm indicates a serious risk of reflux requiring immediate attention.

[0082] The laryngeal mask airway (LMA) is used to fix the gastric tube at the patient's glottis. Its ergonomic design ensures stability during insertion and use. The LMA has pre-drilled insertion holes for the gastric tube, facilitating rapid insertion and precise positioning.

[0083] The above embodiments are merely exemplary embodiments of this utility model and are not intended to limit this utility model. The scope of protection of this utility model is defined by the claims. Those skilled in the art can make various modifications or equivalent substitutions to this utility model within its substance and scope of protection, and such modifications or equivalent substitutions should also be considered to fall within the scope of protection of this utility model.

[0084] The above is merely an illustrative description of the present utility model. Those skilled in the art should know that various improvements can be made to the present utility model without departing from its working principle, and all of these improvements fall within the protection scope of the present utility model.

Claims

1. A reflux recognition device of a smart gastric tube, characterized in that, include: A gastric tube, which is inserted into the patient's esophagus via a laryngeal mask, and is equipped with a video information acquisition device; A video information acquisition device, including a camera and sensor unit, is used to acquire video data inside the patient's esophagus; The data processing module, connected to the video information acquisition device, is used to analyze the video data frame by frame and detect the re-entrant, mark the location of the re-entrant, and generate re-entrant data; An attraction module, connected to the data processing module, is used to select different attraction levels based on the return data; An alarm control module, connected to the data processing module, is used to set alarm status based on the returned data.

2. The reflux recognition device of the smart gastric tube according to claim 1, characterized by, The data processing module specifically includes: The data reading unit is used to read the video data frame by frame; The target detection unit is used to input frame-by-frame images into a pre-trained YOLO image detection model to identify the re-flowing objects and mark their specific locations; The secondary analysis unit is used to perform secondary target extraction on the marked reflux image, analyze the reflux magnitude, and generate reflux data.

3. The reflux identification device for intelligent gastric tubes according to claim 1, characterized in that, The data processing module specifically includes: The multi-frame processing unit is used to select multiple consecutive frames of images and input them into the convolutional LSTM model to predict the state of the reef in future frames of images. The change analysis unit is used to analyze the changing trend and spatial distribution characteristics of the amount of reflux based on the prediction results of the convolutional LSTM model. The visualization unit is used to annotate the predicted bounding box of the return flow and generate visualization results.

4. The reflux identification device for intelligent gastric tubes according to claim 1, characterized in that, Specifically, the data processing module can generate first return data, second return data, or third return data based on the size ratio of the returned material. The attraction module specifically includes: an intensity control unit, used to select a first-level attraction intensity based on the first return flow data, a second-level attraction intensity based on the second return flow data, and a third-level attraction intensity based on the third return flow data.

Images