Warning device for abnormal position of endotracheal tube

Abstract

To provide a warning device for an abnormal position of an endotracheal tube.SOLUTION: A warning device for an abnormal position of an endotracheal tube that monitors accuracy of an installation position of an endotracheal tube includes: a monitoring module 1 for collecting chest X-ray image materials of a patient; an object detection module 2 for detecting an object based on a deep learning model; a position evaluation module 3 for interpreting the position appropriateness of the endotracheal tube based on a result of the object detection module; and a display module 4 for displaying an evaluation result on the position appropriateness of the endotracheal tube and providing a necessary warning as a clinical reference. Thereby, the position of the endotracheal tube is evaluated by an artificial intelligence calculation model using a chest X-ray image captured for the patient. A main objective is to enable clinical medical staff to quickly and accurately determine the appropriateness of the position of the endotracheal tube, reduce the rate of unplanned slipping of the endotracheal tube and the incidence of single-lung ventilation, and improve the medical quality, thereby improving the safety and vital sign stability of the patient.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to an endotracheal tube abnormal position warning device, particularly in medical images, which can help clinical medical staff quickly and accurately confirm whether an endotracheal tube is properly positioned, thereby reducing the incidence of unplanned endotracheal tube slippage and one-lung ventilation, thereby improving medical quality and enhancing patient safety and vital sign stability. [Background technology]

[0002] Although the incidence of unplanned endotracheal tube (ETT) slippage and one-lung ventilation is not high, when it does occur, it can lead to unstable patient vital signs, increased hospital stays and medical costs, and, in the worst case, even death. Therefore, the accuracy of endotracheal tube placement is becoming increasingly important in clinical practice. Furthermore, in the medical environment, for patients in intensive care units (ICUs), an average chest radiograph is taken every two to five days, and these radiographs now require manual interpretation by physicians. Therefore, when clinicians review chest radiographs under a heavy workload, they are constantly being aspirated for major lesions such as pneumonia, and are therefore required to check the position of the endotracheal tube one by one.

[0003] In intensive care units, unplanned endotracheal tube slippage and one-lung ventilation events occasionally occur, resulting in fatal injuries to patients. Ideally, endotracheal tube position should be carefully inspected for each chest radiograph. However, because physicians are very busy and under heavy workloads, they usually focus on other obvious lesions, and thus endotracheal tube displacement may be ignored. Therefore, a device that can detect and judge the position of the endotracheal tube and issue a warning is needed to assist physicians in their judgment. DISCLOSURE OF THE INVENTION [Problem to be solved by the invention]

[0004] The main objective of the present invention is to solve the above-mentioned problems of the conventional technology and provide an artificial intelligence (AI) calculation model that can automatically identify the position of an endotracheal tube in chest X-ray images, thereby enabling clinical medical staff to quickly and accurately confirm whether the endotracheal tube is properly positioned, reducing the incidence of unplanned endotracheal tube slippage and one-lung ventilation, improving medical quality, and providing an endotracheal tube abnormal position warning device that improves patient safety and vital sign stability.

[0005] Another object of the present invention is to provide an artificial intelligence (AI) computational model that can automatically interpret the appropriateness of endotracheal tube positioning in chest X-ray images and issue an appropriate warning, and this AI computational model has excellent performance in both the development and testing stages, and in clinical practice, this model can be used to provide a warning device for endotracheal tube abnormal positioning, which can improve the quality and safety of patient care.

[0006] Another object of the present invention is to take one chest X-ray image of a patient in an intensive care unit every two to five days on average to evaluate the appropriateness of endotracheal tube positioning, and to use multiple chest X-ray images to train an artificial intelligence computation model, which then: 1) determines the position of the tracheal carina on the chest X-ray image to determine whether an endotracheal tube or tracheostomy tube is being used and its position; and 2) calculates the appropriateness of the endotracheal tube position and issues a warning if the endotracheal tube is abnormal.

[0007] Another object of the present invention relates to the field of image processing, and in particular to a method for distance scaling of an image and marking. In image processing applications, the distance between objects in an image is generally calculated and measured for analysis and marking. However, because different images may have different sizes and proportions, directly representing distance using pixel values ​​may be inaccurate. Therefore, a method is needed to convert image pixel values ​​into real-world distance values ​​for accurate image analysis and marking.

[0008] The method includes: step 1: defining an original size of a training dataset image, including the height and width of the image; step 2: receiving an input image to be processed; step 3: calculating a scaling r, which is the ratio between the size of the input image and the size of the training dataset image, and selecting a relatively small value as the scaling; step 4: converting pixel values ​​of the input image into actual distance values ​​corresponding to the training dataset image based on the calculated scaling r; and step 5: using the converted actual distance values ​​for image analysis and marking, such as distance calculation and marking between objects. [Means for solving the problem]

[0009] In order to achieve the above-mentioned objects, the present invention provides an endotracheal tube position abnormality warning device for monitoring the accuracy of the placement position of an endotracheal tube, the device comprising: a monitoring module for collecting image material from a patient's chest X-ray; a device connected to the monitoring module for receiving the image material from the chest X-ray input from the monitoring module; and an artificial intelligence (AI) calculation model for detecting the tracheal carina, endotracheal tube (ETT), and tracheostomy tube on the image. an object detection module that identifies the presence of an object (tracheal carina) or an endotracheal tube and annotates its position; a position evaluation module that is connected to the object detection module and, when the object detection module detects the presence of a tracheal carina or an endotracheal tube, automatically measures the distance between the tip of the endotracheal tube and the tracheal carina based on the annotated position determined by the artificial intelligence computation model, and generates an evaluation result for the appropriateness of the endotracheal tube position based on the determined distance value, and triggers an alarm if the evaluation result shows that the endotracheal tube is not located within an accurate range; and a display module that is connected to the position evaluation module and receives the determination results of the object detection module and the position evaluation module, issues a warning signal, and displays the annotation result of the object and the evaluation result of the appropriateness of the endotracheal tube position.

[0010] According to the above embodiments of the present invention, the object detection module is based on the YOLOv5 deep learning model, inputting chest X-ray image materials, and using a plurality of chest X-ray image materials with a preset ratio as a training dataset and a test dataset for deep learning, respectively, to train the model using the training dataset, and then perform testing using the test dataset. After that, the AI ​​computation model is finalized through verification, and the performance of the AI ​​computation model is evaluated according to the evaluation index of the standard function.

[0011] According to the above embodiments of the present invention, the standard performance evaluation metrics include, but are not limited to, precision, recall, mean average precision (mAP@50), and accuracy.

[0012] According to these embodiments of the present invention, the position assessment module triggers an alarm if the position of the endotracheal tube tip is lower than the tracheal carina or if the distance between the endotracheal tube tip and the tracheal carina is less than 3 cm or greater than 5 cm.

[0013] According to the above embodiments of the present invention, the chest x-ray image material of the object detection module includes chest x-rays with endotracheal tubes, chest x-rays with tracheostomy tubes, and chest x-rays with or without endotracheal tubes.

[0014] According to the above embodiments of the present invention, the source of chest x-rays with endotracheal tubes is primarily images of actual patients requiring endotracheal tube therapy, including cases of proper and improper endotracheal tube placement, and the patient group includes patients with different endotracheal tube experiences and chest anatomies. BEST MODE FOR CARRYING OUT THE INVENTION

[0015] 1 to 4F, there are shown a structural conceptual diagram of an endotracheal tube position abnormality warning device according to the present invention, a conceptual diagram of the usage flow according to the present invention, a conceptual diagram of when an artificial intelligence computation model according to the present invention determines that an endotracheal tube is not present, a conceptual diagram of when an artificial intelligence computation model according to the present invention determines that the position of the endotracheal tube is correct, a conceptual diagram of when an artificial intelligence computation model according to the present invention determines that the position of the endotracheal tube is abnormal, a conceptual diagram of when an artificial intelligence computation model according to the present invention determines that the position of the endotracheal tube is in an abnormal distance range, an image diagram of when the shortest of the three determined distances is taken by the artificial intelligence computation model according to the present invention, and a diagram of when the position of the tracheal carina is detected by the artificial intelligence computation model according to the present invention. 1 is an image showing an endotracheal tube position abnormality warning device that detects the accuracy of endotracheal tube placement, but does not detect an endotracheal tube; an image showing an artificial intelligence computation model of the present invention detecting a tracheal carina and a tracheostomy tube; an image showing an artificial intelligence computation model of the present invention detecting a tracheal carina and an endotracheal tube and determining that the position of the endotracheal tube tip is within a normal distance range; an image showing an artificial intelligence computation model of the present invention detecting a tracheal carina and an endotracheal tube and determining that the distance between the position of the endotracheal tube tip and the tracheal carina is less than 3 cm; and an image showing an artificial intelligence computation model of the present invention detecting a tracheal carina and an endotracheal tube and determining that the distance between the position of the endotracheal tube tip and the tracheal carina is greater than 5 cm. As shown in the figures, the present invention is an endotracheal tube position abnormality warning device that monitors the accuracy of endotracheal tube placement, and is composed of a monitoring module 1, an object detection module 2, a position evaluation module 3, and a display module 4.

[0016] The monitoring module 1 collects image material from a patient's chest x-ray.

[0017] The property detection module 2 is connected to the monitoring module 1 and detects properties based on the YOLO V5 deep learning model.

[0018] The position evaluation module 3 is connected to the object detection module 2 and determines the appropriateness of the position of the endotracheal tube based on the result of the object detection module 2.

[0019] The display module 4 is connected to the position assessment module 3, and displays the assessment result of the proper positioning of the endotracheal tube, and issues a warning as necessary for clinical reference. The above structure constitutes a novel endotracheal tube position abnormality warning device 100.

[0020] The distance measurement is performed by a position distance appropriateness calculation module. The source code is inspected to detect at least two objects, with the first object being a tracheal carina and the second an endotracheal tube, or vice versa. If a tracheal carina or endotracheal tube is detected, the source code reads the corresponding coordinate information from the detected object. The source code then calculates the distance between the endotracheal tube and the tracheal carina. The calculation process involves: 1) using the Euclidean distance formula to calculate the linear distance between the two coordinate objects; 2) scaling the predicted image and the training dataset image using the calculated distance value, while simultaneously converting the pixel values ​​into actual distance values. For the position of the endotracheal tube tip, the leftmost and rightmost coordinate points are selected from the bottom edge of the annotation bounding box of the identified endotracheal tube tip. Then, the midpoint between the two points is calculated using a function. For example, points A and C in Figure 4A are the rightmost and leftmost points at the bottom of the bounding box, and point B is the midpoint at the bottom of the bounding box, i.e., the midpoint between points A and C. The distance values ​​between the three points A, B, and C and the tracheal carina (e.g., point D in Figure 4A) are calculated, and the shortest distance is taken as the result. The relevant coordinate points and distance marks, along with the distance values ​​and warnings, are displayed on the image.

[0021] 4B to 4F show examples of results determined by the artificial intelligence computation model according to the present invention. In FIG. 4B, the model detects the position of the tracheal carina but does not detect the endotracheal tube. FIG. 4C shows that the model detects the tracheal carina or tracheotomy tube. FIG. 4D shows that the model detects the tracheal carina or endotracheal tube and the tip position of the endotracheal tube is within the normal distance range. FIG. 4E shows that the model detects the tracheal carina or endotracheal tube and the distance between the tip position of the endotracheal tube and the tracheal carina is less than 3 cm. FIG. 4F shows that the model detects the tracheal carina or endotracheal tube and the distance between the tip position of the endotracheal tube and the tracheal carina is greater than 5 cm.

[0022] According to the present invention, the object detection module 2 uses an artificial intelligence (AI) computational model to identify and annotate the tracheal carina, endotracheal tube (ETT), and tracheostomy tube in the chest X-ray image data input from the monitoring module 1. When the object detection module 2 detects the presence of a tracheal carina or endotracheal tube, the position evaluation module 3 automatically measures the distance between the tip of the endotracheal tube and the tracheal carina based on the annotation position determined by the AI ​​computational model. Based on the distance value, the module generates an evaluation result of the appropriateness of the endotracheal tube position. If the evaluation result indicates that the endotracheal tube is not within the correct range, an alarm is triggered. Finally, the display module 4 receives the results of the object detection module 2 and the position evaluation module 3, issues an alarm, and displays the object annotation result and the evaluation result of the appropriateness of the endotracheal tube position. Therefore, the device 100 can be widely applied in the medical industry, and in order to reduce the workload of clinicians, the present invention uses artificial intelligence technology to detect and determine the position of the endotracheal tube, thereby providing important reference information for the doctor's judgment on the appropriateness of the tracheal position.

[0023] According to the present invention, the object detection module 2 and position assessment module 3 for assessing the appropriateness of endotracheal tube position use chest X-ray images taken of the patient and an advanced artificial intelligence computation model to assess the position of the endotracheal tube. The main purpose is to enable clinical medical staff to quickly and accurately determine whether the endotracheal tube is properly positioned, thereby reducing the incidence of unplanned endotracheal tube slippage and one-lung ventilation, thereby improving the quality of medical care, patient safety, and vital sign stability.

[0024] According to the present invention, the object detection module 2 uses 2,278 sample images to train the YOLO V5 model, and then tests and validates using 253 images. The presence of an endotracheal tube or tracheostomy tube is confirmed by predicting the tracheal carina position in the X-ray image. Once an endotracheal tube is detected, the position assessment module 3 further calculates the distance between the endotracheal tube tip and the tracheal carina, eliminating the problem of the endotracheal tube being placed too deep. Standard performance evaluation metrics include, but are not limited to, precision, recall, mean average precision (mAP@50), and accuracy, ensuring the high accuracy of the model.

[0025] The above distance scaling formula calculates the ratio between the pixel values ​​of the input image and the predicted image and the training dataset image, and then converts them into distance values ​​for practical applications, thereby achieving accurate image analysis and marking. This method is simple and easy to implement, so it can be widely used in image processing applications, improving the accuracy and efficiency of image analysis.

[0026] According to the present invention, the object detection module 2 and the position assessment module 3 do not require any special equipment. Moreover, they are compatible with existing chest X-ray image acquisition equipment and medical systems, providing a simple integrated solution. Clinical staff only need to input the patient's X-ray image material into the information system, and the assessment result of the endotracheal tube position can be obtained quickly.

[0027] According to a more specific embodiment of the present invention, the image materials of the chest X-ray of the object detection module 2 include chest X-rays with an endotracheal tube, chest X-rays with a tracheostomy tube, and chest X-rays without an endotracheal tube or a tracheostomy tube.

[0028] According to a more specific embodiment of the present invention, the source of the chest X-ray with endotracheal tube is mainly images of actual patients who require endotracheal tube treatment, and includes examples of proper and improper endotracheal tube placement, and the patient group includes patients with different endotracheal tube experiences and chest anatomies.

[0029] According to a more specific embodiment of the present invention, the position assessment module 3 triggers an alarm when the position of the tip of the endotracheal tube is lower than the tracheal carina, or when the distance between the tip of the endotracheal tube and the tracheal carina is less than 3 cm or greater than 5 cm.

[0030] The following examples are given to illustrate the details and meaning of the present invention, but the scope of the claims of the present invention is not limited thereby.

[0031] In a preferred embodiment, the source of test material is actual chest x-ray images of patients. Normal and incorrect endotracheal tube placements are included. A variety of different cases and patients are included. Data is anonymized to protect patient privacy and compliance. Patients with different endotracheal tube experiences and chest anatomies are included.

[0032] The total number of chest radiographs in the test was 253.

[0033] The test results are (1) Sensitivity is 0.963, (2) Specificity is 0.964, (3) Mean average precision (mAP@50): 0.966 (4) Accuracy is 0.962.

[0034] According to a preferred embodiment, the flow is as shown in FIG.

[0035] In step s11, a pre-trained YOLOv5 model is used to detect the image material of a chest X-ray and identify the location of the tracheal carina, endotracheal tube, and tracheostomy tube. As shown in Figure 4B, the location of the tracheal carina is detected, but the endotracheal tube is not present in the image.

[0036] In step s12 of automatically identifying the tip position of the endotracheal tube and automatically measuring the distance between the tip of the endotracheal tube and the tracheal carina, the distance between the tip of the endotracheal tube and the tracheal carina is automatically measured using a model.

[0037] Step s13 of determining whether the position of the endotracheal tube is correct determines whether the position of the endotracheal tube is appropriate based on the position of the tip of the endotracheal tube and the measured distance between the tip of the endotracheal tube and the tracheal carina, as shown in Figure 3B.

[0038] In step s14 of issuing an early warning, if the tip position of the endotracheal tube is lower than the tracheal carina, or if the measured distance is not within the range of 3 to 5 cm, as shown in Figure 3C, a warning is issued that the endotracheal tube may be positioned incorrectly.

[0039] In Figures 4B-4F, an artificial intelligence computational model identifies the presence and location of the tracheal carina, endotracheal tube, and tracheostomy tube in the image and measures the distance between the endotracheal tube tip and the tracheal carina. At the same time, referring to past clinical standards and literature, a standard rule for triggering an endotracheal tube alarm is proposed based on the standard. According to this rule, as shown in Figure 3B, a green light is turned on when the distance is 3-5 cm, as shown in Figure 3D, a yellow light is turned on when the distance is 2.5-3 cm or 5-5.5 cm, as shown in Figure 3C, and a red light is turned on when the distance is less than 2.5 cm or more than 5.5 cm. In another simplified embodiment, as shown in Figure 3C, a red light is turned on when the distance is less than 3 cm or more than 5 cm. Furthermore, when only the tracheal carina is detected and no endotracheal tube is detected, the display shown in Figure 3A is displayed. In a preferred embodiment, if an alarm is first sounded, a visual flash or sound will be displayed to reinforce the attention, and once the clinical staff is aware of the alarm, the visual flash or sound will stop.

[0040] As can be seen from the above, for critically ill patients receiving mechanical ventilation, unplanned extubation or one-lung ventilation occasionally occur, potentially resulting in fatal injuries. In theory, the position of the endotracheal tube on each chest X-ray image should be carefully examined. However, endotracheal tube displacement can be easily overlooked by physicians, who may be drawn to other obvious lesions. Therefore, this invention develops an artificial intelligence (AI) computational model that can automatically identify the position of the endotracheal tube on chest X-ray images. The materials and methods used are to randomly select anonymized chest X-ray images of critically ill patients to establish training and test datasets, and to establish an AI computational model using the Python YOLOv5 model. This model can predict the position of the tracheal carina on X-ray images and identify the presence of an endotracheal tube or tracheostomy tube. If an endotracheal tube is detected, the distance between the tracheal carina and the tip of the endotracheal tube is calculated. An alarm is triggered when the tip of the endotracheal tube is located below the tracheal carina or when the distance is less than 3 cm or greater than 5 cm. According to the present invention, based on experimental results, the precision, recall, mean average accuracy, and accuracy of an AI computational model using 2,278 chest X-ray images as a training dataset and 253 chest X-ray images as a test dataset were 0.963, 0.964, 0.966, and 0.962, respectively. The present invention also implements the AI ​​computational model in clinical practice as external validation. The alarm system triggered by the AI ​​computational model showed a median (interquartile range) reduction in the duration of improper endotracheal tube placement from 3.00 (1.25-4.00) days to 2.00 (1.00-3.00) days. Therefore, the present invention proposes an AI computational model that can automatically interpret the appropriateness of endotracheal tube placement in chest X-ray images and issue appropriate warnings. Furthermore, the AI ​​computational model achieved excellent performance during both the development and testing phases. Applying the model to clinical practice can improve the quality and safety of patient care.

[0041] The present invention has application in medical imaging, and its primary purpose is to enable clinical staff to quickly and accurately determine whether an endotracheal tube is properly positioned, thereby reducing the incidence of unplanned endotracheal tube slippage and one-lung ventilation, and improving the stability of a patient's vital signs.

[0042] 1. In the above-mentioned endotracheal tube position abnormality warning device of the present invention, the artificial intelligence calculation module for evaluating the appropriateness of endotracheal tube positioning has the following applications and effects:

[0043] 1. Automatic evaluation of endotracheal tube position on patient chest X-ray images. 2. Help clinical medical staff quickly and accurately determine whether endotracheal tube placement is appropriate. 3. Reduce the incidence of unplanned endotracheal tube slippage and one-lung ventilation, lower the risk of related complications, and improve medical quality, patient safety, and vital sign stability.

[0044] Second, the indications for use of the endotracheal tube abnormal position warning device according to the present invention are as follows:

[0045] 1. The indications for using the above-mentioned artificial intelligence calculation module include, but are not limited to, using the chest X-ray images taken by the patient to determine whether the endotracheal tube is properly positioned. 2. Applies to all patients requiring endotracheal tube therapy, including but not limited to critically ill patients and surgical patients.

[0046] On average, a patient in an intensive care unit receives a chest X-ray every two to five days, which is then manually interpreted by a physician. However, under the enormous workload of clinicians, when reviewing chest X-rays, their attention is often drawn to major pathologies such as pneumonia, and they are unable to check the position of each endotracheal tube one by one. Therefore, the present invention provides an endotracheal tube abnormal position warning device that can identify and detect the position of a patient's endotracheal tube and provide instant evaluation results, primarily as an auxiliary tool for providing endotracheal tube warnings to clinical medical staff, thereby ensuring proper endotracheal tube positioning, improving patient treatment outcomes, and reducing associated risks.

[0047] As described above, the technical features of the present invention are as follows: 1. Train with the YOLO V5 model. A sample of 2278 images is used for training, and 253 images are used for testing and validation. 2. Predict X-ray images. a. While checking the position of the tracheal carina, check for the presence of an endotracheal tube or tracheostomy tube. b. If an endotracheal tube is present, the distance between the tracheal carina and the tip of the endotracheal tube is further calculated to determine the appropriateness of the endotracheal tube position. 3. It is a proof of concept.

[0048] As described above, the present invention provides an endotracheal tube position abnormality warning device that effectively overcomes the drawbacks of the prior art. For patients in an intensive care unit, one chest X-ray is taken on average every two to five days to evaluate the appropriateness of endotracheal tube positioning. A number of chest X-ray image samples are used to train an artificial intelligence (AI) computational model. The AI ​​computational model then: 1) determines the position of the tracheal carina on the chest X-ray image and whether an endotracheal tube (ETT) or tracheostomy tube is used, and then determines its position; and 2) calculates the appropriateness of endotracheal tube positioning and issues a warning. Therefore, the present invention is more advanced and practical, and can certainly meet users' needs. Therefore, the following patent claims are submitted in accordance with the law.

[0049] The above description is merely a preferred embodiment of the present invention, and the scope of the present invention is not limited thereby, and various equivalent changes and modifications are included in the scope of the claims of the present invention. [Brief explanation of the drawings]

[0050] [Figure 1] 1 is a structural conceptual diagram of an endotracheal tube abnormal position warning device according to the present invention. [Figure 2] FIG. 1 is a conceptual diagram of a usage flow according to the present invention. [Figure 3A] FIG. 1 is a conceptual diagram of the artificial intelligence calculation model according to the present invention when determining that an endotracheal tube is not present. [Figure 3B] FIG. 1 is a conceptual diagram of the artificial intelligence calculation model according to the present invention determining that the endotracheal tube position is correct. [Figure 3C] FIG. 10 is a conceptual diagram of the artificial intelligence computation model according to the present invention when determining that the endotracheal tube position is abnormal. [Figure 3D] FIG. 10 is a conceptual diagram of when the artificial intelligence computation model according to the present invention determines that the endotracheal tube position is almost abnormal. [Figure 4A]FIG. 10 is an image diagram showing the shortest distance among three distances determined by the artificial intelligence calculation model according to the present invention. [Figure 4B] FIG. 10 is an image diagram showing the state when the artificial intelligence computation model according to the present invention detects the position of the tracheal carina but does not detect the endotracheal tube. [Figure 4C] FIG. 10 is an image diagram showing the tracheal carina and tracheostomy tube detected by the artificial intelligence computation model according to the present invention. [Figure 4D] 10 is an image diagram showing an image when the artificial intelligence calculation model according to the present invention detects the tracheal carina and the endotracheal tube and determines that the tip position of the endotracheal tube is within the normal distance range. FIG. [Figure 4E] This is an image diagram of an artificial intelligence computation model according to the present invention detecting the tracheal carina and endotracheal tube and determining that the distance between the tip of the endotracheal tube and the tracheal carina is less than 3 cm. [Figure 4F] This is an image diagram of an artificial intelligence computation model according to the present invention detecting the tracheal carina and endotracheal tube and determining that the distance between the tip of the endotracheal tube and the tracheal carina is greater than 5 cm. [Explanation of symbols]

[0051] 1. Monitoring module Property detection module 100 Endotracheal tube misplacement warning device 2. Location Evaluation Module 3 Display Module 4 Steps s11~s14

Claims

1. The system comprises a monitoring module that monitors the accuracy of endotracheal tube placement and collects image material from the patient's chest x-ray; an object detection module that is connected to the monitoring module and receives the image material from the chest x-ray input from the monitoring module, and uses an artificial intelligence (AI) computational model to identify the presence and annotate the location of the tracheal carina, endotracheal tube (ETT), and tracheostomy tubes in the image; a position evaluation module connected to the object detection module, which, when the object detection module detects the presence of a tracheal carina and an endotracheal tube, automatically measures the distance between the tip of the endotracheal tube and the tracheal carina using the annotation position determined by the artificial intelligence calculation model, generates an evaluation result for the appropriateness of the endotracheal tube position based on the obtained distance value, and triggers an alarm when the endotracheal tube is not positioned within an accurate range; and a display module connected to the position evaluation module, which receives the determination results of the object detection module and the position evaluation module, issues a warning signal, and displays the object annotation result and the evaluation result for the appropriateness of the endotracheal tube position.

2. The object detection module inputs chest X-ray image materials based on the YOLOv5 deep learning model, and uses a plurality of chest X-ray image materials with a preset ratio as a training dataset and a test dataset for deep learning, respectively, to train the model using the training dataset, and then tests and confirms using the test dataset to complete the artificial intelligence calculation model, and then evaluates the performance of the artificial intelligence calculation model using standard function evaluation indexes, which is the endotracheal tube abnormal position warning device described in claim 1.

3. The endotracheal tube abnormal position warning device of claim 2, wherein the standard performance evaluation indexes include, but are not limited to, precision, recall, mean average precision (mAP@50), and accuracy.

4. The position assessment module measures the distance between the endotracheal tube tip and the tracheal carina, selects the coordinate points of the left and right ends of the bottom edge of the annotation bounding box of the identified endotracheal tube tip, calculates the midpoint between the two points using a function, and selects the shortest calculated distance using the distance values ​​between the three points and the tracheal carina using the Euclidean distance formula. Furthermore, the pixel values ​​of the image are converted into corresponding actual distance values ​​for the distance.

5. The endotracheal tube position abnormality warning device of claim 1, wherein the position assessment module triggers an alarm when the position of the endotracheal tube tip is lower than the tracheal carina, or when the distance between the endotracheal tube tip and the tracheal carina is less than 3 cm or greater than 5 cm.

6. The endotracheal tube abnormal position warning device according to claim 1 or 5, characterized in that the alarm triggering method is represented by, but not limited to, a visual, flashing visual, or sound.

7. The endotracheal tube abnormal position warning device according to claim 1, wherein the chest X-ray image materials of the object detection module are chest X-rays with an endotracheal tube, chest X-rays with a tracheostomy tube, and chest X-rays without an endotracheal tube or a tracheostomy tube.

8. The endotracheal tube abnormal position warning device of claim 7, characterized in that the source of the chest X-ray with endotracheal tube is mainly images of actual patients who require endotracheal tube treatment, which includes cases of whether the endotracheal tube is properly positioned or not, and the patient group includes patients with different endotracheal tube experiences and chest anatomical structures.

Images