Airway opening body position included angle recommendation method based on intelligent model prediction

Abstract

The invention discloses an airway opening body position included angle recommendation method based on intelligent model prediction, which comprises the following steps: deploying a random forest for predicting a body position included angle into an application program, and receiving the random forest accessed by a user through a graphical user interface of the random forest application program, and the predicted body position included angle obtained by the random forest is fed back to the user. The prediction model provided by the invention can accurately recommend the optimal glottis-exposed airway opening body position for medical personnel, so that the risk of difficult glottis exposure in bronchoscope diagnosis and treatment can be reduced or avoided, and time can be saved for bronchoscope diagnosis and treatment; and a prediction service is provided through a computer application program, so that the use convenience is good.

Description

Technical Field

[0001] The present invention relates to the field of medical technology, and particularly to a method for recommending the optimal airway opening angle. Background Art

[0002] Glottis exposure refers to the situation where the glottis structure cannot be smoothly observed due to various reasons during tracheal intubation or laryngoscopy. According to domestic and foreign research, the incidence of difficult glottis exposure under laryngoscope is 1.5% - 13.0%. Due to the limited glottis view, it often increases the risks of airway mucosal injury and aspiration in patients, and may even lead to failed intubation or prolonged intubation time, resulting in severe hypoxemia in patients, causing irreversible brain damage or death.

[0003] Fully and effectively opening the airway to achieve the best glottis exposure is an effective measure to reduce discomfort and complications after bronchoscopy. The inventor previously disclosed a control method for an airway opening position automatic adjustment device in a patent with the application number 2022111651871. Through this method in the experiment of opening the patient's airway, it was found that when the airway was opened at the three body position included angles of 90°, 95°, and 100°, the CL (Cormack - Lehane) grading of the vast majority of patients was grade I or grade II, indicating that this control method can effectively open the patient's airway and obtain a good glottis view, which is beneficial for the bronchoscope to smoothly pass through the glottis into the lower respiratory tract and continuously maintain the patency of the airway.

[0004] However, due to individual differences among patients, when opening the patient's airway, which of the included angles of 90°, 95°, and 100° is the optimal included angle for glottis exposure of this patient is unknown to medical staff. If each angle is tried once, it will consume precious treatment time. Summary of the Invention

[0005] In view of this, the purpose of the present invention is to provide a method for recommending the included angle of the airway opening position based on intelligent model prediction to solve the technical problem of helping medical staff know in advance the best included angle of the airway opening position for glottis exposure of patients.

[0006] The method for recommending the included angle of the airway opening position based on intelligent model prediction of the present invention includes deploying a random forest for predicting the included angle of the position to an application program, receiving user access to the random forest through the graphical user interface of the application program, and feeding back the predicted included angle of the position obtained by the random forest to the user;

[0007] The random forest is obtained through the following training and verification:

[0008] 1) Collect the independent variable data affecting the glottis exposure effect when opening the patient's airway, and each patient corresponds to a set of independent variable data;

[0009] 2) Conduct airway opening experiments on the patients corresponding to each set of independent variable data. During the experiments, collect the glottis exposure images of the patients at three preselected body position angles of 90°, 95°, and 100° respectively. Perform laryngoscope exposure grading on the glottis exposure status in the images, and select the optimal body position angle that enables the patient to be in the best glottis exposure state from the three preselected body position angles according to the laryngoscope exposure grading situation; and save the optimal body position angle and its corresponding independent variable data.

[0010] 3) Perform assignment processing on the count independent variable data in each set of saved independent variable data, retain the original values of the measurement independent variable data in each set of saved independent variable data, perform assignment processing on the optimal body position angles corresponding to each set of saved independent variable data, and the assignment of the optimal body position angle serves as the outcome dependent variable; construct a data set with each set of independent variable data and the corresponding outcome dependent variable data after the above-mentioned processing, divide the data set into a training set and a validation set, and use the data in the training set and the validation set to train and validate the random forest respectively to obtain a random forest for predicting the body position angle.

[0011] Further, in step 2), the method for selecting the optimal body position angle that enables the patient to be in the best glottis exposure state from the three preselected body position angles according to the laryngoscope exposure grading situation includes:

[0012] According to the laryngoscope exposure grading of the glottis exposure, if there is a preselected body position angle with a laryngoscope exposure grading below grade II, then use the preselected body position angle with the lowest laryngoscope exposure grading as the optimal body position angle. If there are more than two of the aforementioned preselected body position angles, select the one with the smallest angle value as the optimal body position angle;

[0013] If the laryngoscope exposure gradings corresponding to the three preselected body position angles are all above grade III, then there is no corresponding optimal body position angle for this set of independent variable data, and this set of independent variable data is deleted.

[0014] Further, the independent variable data collected in step 1) includes: age, modified Mallampati score, hyoid-mental distance, thyroid-mental distance, sternal-mental distance, ratio of height to thyroid-mental distance, ratio of neck circumference to thyroid-mental distance, and ratio of height to hyoid-mental distance;

[0015] The hyoid-mental distance refers to the distance from the anterior edge of the hyoid bone to the mental protuberance of the mandible;

[0016] The thyroid-mental distance refers to the straight-line distance from the superior notch of the thyroid cartilage to the mental angle of the mandible when the patient is in the maximum head extension state with the mouth closed;

[0017] The sternal-mental distance refers to the straight-line distance from the suprasternal notch to the mental angle of the mandible when the patient has the mouth closed and the head and neck are in the maximum extension position;

[0018] The height-mental distance ratio refers to the ratio of height to mental distance;

[0019] The neck circumference-mental distance ratio refers to the ratio of neck circumference to mental distance;

[0020] The height-tongue-mental distance ratio refers to the ratio of height to tongue-mental distance.

[0021] Further, the method for obtaining the modified Mallampati score is as follows: The patient takes a sitting position, with the head in a neutral position, opens the mouth as wide as possible and protrudes the tongue to the maximum extent without making a sound, and is graded according to the visibility of the oropharyngeal structure; Grade I: The soft palate, pharyngopalatine arch, and uvula can be seen; Grade II: The soft palate and pharyngopalatine arch can be seen, but part of the uvula is blocked by the root of the tongue; Grade III: Only the soft palate can be seen; Grade IV: The soft palate cannot be seen.

[0022] Further, the application program is a web application program or a desktop application program.

[0023] Further, the random forest is deployed to the web application program by using Flask.

[0024] Advantages of the present invention:

[0025] 1. The method for recommending the airway opening body position angle based on intelligent model prediction of the present invention provides a prediction model that can accurately recommend the optimal body position angle for medical staff to open the patient's airway and expose the glottis, thereby reducing or avoiding the risk of difficult glottis exposure during bronchoscopy diagnosis and treatment. By knowing in advance the optimal body position angle for opening the patient's airway and exposing the glottis, it can also save time for bronchoscopy diagnosis and treatment.

[0026] 2. The method for recommending the airway opening body position angle based on intelligent model prediction of the present invention deploys the random forest for predicting the body position angle in a web application program, enabling medical staff to remotely use the random forest for prediction through the network. The random forest can also be deployed in a desktop application program, so that in the absence of a network, medical staff can also perform body position angle prediction through the random forest desktop application program installed on a personal computer. Description of the Drawings

[0027] Figure 1 It is a graphical user interface for the random forest deployed in a web application program. Detailed Embodiments

[0028] The present invention will be further described below with reference to the drawings and embodiments.

[0029] As shown in the figure, the method for recommending the airway opening body position angle based on intelligent model prediction in this embodiment includes deploying a random forest for predicting the body position angle into an application program, receiving user access to the random forest through the graphical user interface of the application program, and feeding back the predicted body position angle obtained by the random forest to the user.

[0030] The random forest is obtained through the following training and verification:

[0031] 1) Collect various independent variable data that affect the glottis exposure effect when the patient's airway is opened. Each patient corresponds to a set of independent variable data.

[0032] The independent variable data collected in this step includes: age, modified Mallampati score, hyoid-mental distance, thyroid-mental distance, sternal-mental distance, ratio of height to thyroid-mental distance, ratio of neck circumference to thyroid-mental distance, and ratio of height to hyoid-mental distance.

[0033] The method for obtaining the modified Mallampati score is as follows: The patient takes a sitting position, the head is in a neutral position, opens the mouth as wide as possible and extends the tongue to the maximum extent without making a sound, and is graded according to the visibility of the oropharyngeal structure; Grade Ⅰ: The soft palate, pharyngeal palatine arch, and uvula can be seen; Grade Ⅱ: The soft palate and pharyngeal palatine arch can be seen, but part of the uvula is blocked by the root of the tongue; Grade Ⅲ: Only the soft palate can be seen; Grade Ⅳ: The soft palate cannot be seen.

[0034] The hyoid-mental distance refers to the distance from the anterior edge of the hyoid bone to the mental protuberance of the mandible.

[0035] The thyroid-mental distance refers to the straight-line distance from the upper notch of the thyroid cartilage to the mental angle of the mandible when the patient is in the maximum head extension state with the mouth closed.

[0036] The sternal-mental distance refers to the straight-line distance from the suprasternal notch to the mental angle of the mandible when the patient has the mouth closed and the head and neck are in the maximum extension position.

[0037] The ratio of height to thyroid-mental distance refers to the ratio of height to thyroid-mental distance.

[0038] The ratio of neck circumference to thyroid-mental distance refers to the ratio of neck circumference to thyroid-mental distance.

[0039] The ratio of height to hyoid-mental distance refers to the ratio of height to hyoid-mental distance.

[0040] 2) Conduct an airway opening experiment on the patients corresponding to each set of independent variable data. During the experiment, collect the glottis exposure images of the patients at three preselected body position angles of 90°, 95°, and 100° respectively, perform laryngoscopic exposure grading (Cormack-Lehane grade, CL grading) on the glottis exposure state in the images, and select the optimal body position angle that enables the patient to be in the best glottis exposure state from the three preselected body position angles; and save the optimal body position angle and its corresponding independent variable data.

[0041] The Cormack-Lehane grade (CL grade) is as follows:

[0042] Level 1 means that the glottis is completely exposed, and the front and back ends of the glottic fissure can be seen (that is, the anterior commissure and posterior commissure can be seen); Level 2 means that the glottis is partially exposed, and only the back end of the glottic fissure can be seen; Level 3 means that only the tip of the epiglottis or the epiglottis can be seen, but the glottis cannot be exposed; Level 4 means that neither the glottis nor the epiglottis can be exposed.

[0043] Methods for selecting the optimal body angle for optimal glottal exposure from three pre-selected body angles according to the laryngoscopic exposure grade include:

[0044] According to the laryngoscopic exposure grade of glottal exposure, if there is a pre-selected body position angle with a laryngoscopic exposure grade below level 2, the pre-selected body position angle with the lowest laryngoscopic exposure grade is used as the optimal body position angle; if there are more than two pre-selected body position angles, the one with the smallest angle value is selected as the optimal body position angle.

[0045] 3) Assigning values ​​to the counting independent variable data in each saved group of independent variable data, retaining the original values ​​of the measurement independent variable data in each saved group of independent variable data, assigning values ​​to the optimal body position angle corresponding to each saved group of independent variable data, and using the assigned value of the optimal body position angle as the outcome dependent variable; constructing a data set using the independent variable data in each group processed as above and the corresponding outcome dependent variable data, dividing the data set into a training set and a validation set, respectively training and validating the random forest using the data in the training set and the validation set, and obtaining a random forest for predicting the body position angle.

[0046] The following table is an assignment table for count independent variable data and outcome dependent variable data:

[0047]

[0048] The following table is a table of baseline data of the research subjects in this embodiment:

[0049]

[0050] In this embodiment, the random forest is specifically deployed to a web application using Flask. On the graphical user interface of the web application, there are input boxes for respectively inputting independent variables such as age, modified Mallampati score, mento - hyoid distance, thyro - mental distance, sterno - mental distance, ratio of height to thyro - mental distance, ratio of neck circumference to thyro - mental distance, and height to mento - hyoid distance. Medical staff can remotely request and call the random forest through the network protocol HTTP, input the independent variable data into the graphical user interface of the web application, thereby implementing the body position, and then run the random forest to obtain the body position angle predicted by the random forest. The following introduces the specific process of creating a web application of the random forest using the Flask framework:

[0051] The first step: Use joblib to save and load the previously trained optimal algorithm model - the random forest.

[0052] The second step: Create a Flask application to load the random forest and handle prediction requests.

[0053] The third step: Define a list of feature names (including age, modified Mallampati score, mento - hyoid distance, thyro - mental distance, sterno - mental distance, ratio of height to thyro - mental distance, ratio of neck circumference to thyro - mental distance, and height to mento - hyoid distance, and these features will be used for model prediction).

[0054] The fourth step: Define the root route. The "root route" refers to the route associated with the root URL of the application (that is, the base address of the website). It is the route where the page that users first see when accessing the website is located. When users enter the domain name or IP address of the website in the browser and do not specify a specific path, they will default to accessing the page corresponding to the root route.

[0055] The fifth step: Define the prediction route. Defining the prediction route means commanding the Flask application how to obtain user input, process data, call the model for prediction, and how to return the prediction result to the user when receiving a specific POST request, and at the same time, handle possible error situations. In this step, the @app.route decorator is used to bind the / predict path to the predict function and only accept POST requests (POST requests are usually used to submit data to the server. For example, when users enter information in a web form and submit it, the data will be sent to the server in the form of a POST request). The functions of the predict function include:

[0056] · Traverse feature_names (the list of feature names) to obtain the value of each feature from the form data of the POST request.

[0057] · Check if there are missing feature values. If missing, return an error message.

[0058] · Try to convert each eigenvalue to a floating-point number, and return an error message if the conversion fails.

[0059] · Convert the input data to the DataFrame format.

[0060] · Use the loaded rfc (random forest) to predict the input data and obtain the prediction result y_pred.

[0061] · Use the predict_proba method to obtain the prediction probability y_pred_proba for each class.

[0062] · Assume the class labels are [1, 2, 3], and correspond the class labels with the prediction probabilities, storing them in the prediction_probabilities dictionary. Label 1 represents a predicted body position angle of 90°, label 2 represents a predicted body position angle of 95°, and label 3 represents a predicted body position angle of 100°.

[0063] · Return the prediction result and the prediction probability to the client in JSON format.

[0064] · If an exception occurs during the processing, return an error message.

[0065] Through the above process, a web application of a random forest is created, allowing users to input eigenvalue through an HTML form, then use the loaded random forest to predict the body position angle, and return the prediction result and the prediction probability.

[0066] Of course, in different embodiments, the random forest is also deployed to a desktop application, so that in the absence of a network, medical staff can also use the random forest desktop application loaded on a personal computer to predict the body position angle.

[0067] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered within the scope of the claims of the present invention.

Claims

1. A method for recommending the angle of the airway-opening position based on intelligent model prediction, characterized in that: Including deploying a random forest for predicting the body position angle into an application program, receiving user access to the random forest through the graphical user interface of the application program, and feeding back the predicted body position angle obtained by the random forest to the user; The random forest is obtained through the following training and verification: 1) Collect independent variable data that affects the glottis exposure effect when the patient's airway is opened, with each patient corresponding to a set of independent variable data; 2) Conduct an airway opening experiment on the patients corresponding to each set of independent variable data. During the experiment, collect the glottis exposure images of the patients at three preselected body position angles of 90°, 95°, and 100° respectively. Perform laryngoscope exposure grading on the glottis exposure status in the images, and select the optimal body position angle that enables the patient to be in the best glottis exposure state from the three preselected body position angles according to the laryngoscope exposure grading situation; and save the optimal body position angle and its corresponding independent variable data; 3) Perform assignment processing on the count independent variable data in each set of saved independent variable data, retain the original values of the measurement independent variable data in each set of saved independent variable data, perform assignment processing on the optimal body position angle corresponding to each set of saved independent variable data, and the assignment of the optimal body position angle is used as the outcome dependent variable; construct a data set with the processed independent variable data of each group and the corresponding outcome dependent variable data, divide the data set into a training set and a validation set, and use the data in the training set and the validation set to train and verify the random forest respectively to obtain a random forest for predicting the body position angle.

2. The method for recommending the airway opening body position angle based on intelligent model prediction according to claim 1, wherein: In step 2), the method for selecting the optimal body position angle that enables the patient to be in the best glottis exposure state from the three preselected body position angles according to the laryngoscope exposure grading situation includes: According to the laryngoscope exposure grading situation of the glottis exposure, if there is a preselected body position angle with a laryngoscope exposure grading below grade II, then use the preselected body position angle with the lowest laryngoscope exposure grading as the optimal body position angle. If there are more than two of the aforementioned preselected body position angles, select the one with the smallest angle value as the optimal body position angle; If the laryngoscope exposure grades corresponding to the three preselected body position angles are all above grade III, then there is no corresponding optimal body position angle for this set of independent variable data, and this set of independent variable data is deleted.

3. The airway opening body position angle recommendation method based on intelligent model prediction according to claim 1, characterized in that: The independent variable data collected in step 1) includes: age, modified Mallampati score, tongue-mental distance, thyroid-mental distance, sternum-mental distance, height-thyroid-mental distance ratio, neck circumference-thyroid-mental distance ratio, and height-tongue-mental distance; The tongue-mental distance refers to the distance from the anterior edge of the hyoid bone to the mental protuberance of the mandible; The thyroid-mental distance refers to the straight-line distance from the upper notch of the thyroid cartilage to the mental angle of the mandible when the patient is in the maximum head extension state and the mouth is closed; The sternum-mental distance refers to the straight-line distance from the suprasternal notch to the mental angle of the mandible when the patient's mouth is closed and the head and neck are in the maximum extension position; The height-thyroid-mental distance ratio refers to the ratio of height to the thyroid-mental distance; The neck circumference-thyroid-mental distance ratio refers to the ratio of neck circumference to the thyroid-mental distance; The height-tongue-mental distance ratio refers to the ratio of height to the tongue-mental distance.

4. The airway opening body position angle recommendation method based on intelligent model prediction according to claim 3, wherein: The method for obtaining the improved Mallampati score is as follows: The patient takes a sitting position with the head in a neutral position, opens the mouth as wide as possible and protrudes the tongue to the maximum extent without making a sound, and is graded according to the visibility of the oropharyngeal structure; Grade I: The soft palate, pharyngeal palatine arch, and uvula can be seen; Grade II: The soft palate and pharyngeal palatine arch can be seen, but part of the uvula is blocked by the root of the tongue; Grade III: Only the soft palate can be seen; Grade IV: The soft palate cannot be seen.

5. The method for recommending the airway opening body position angle based on intelligent model prediction according to claim 1, wherein: The application program is a web application program or a desktop application program.

6. The method for recommending the airway opening body position angle based on intelligent model prediction according to claim 5, wherein: The random forest is deployed to the web application using Flask.

Images