Artificial intelligence-based difficult airway prediction system and method

Abstract

The invention relates to the technical field of physiological data processing, in particular to a difficult airway prediction system and method based on artificial intelligence. According to the system, the frequency of indirect features of a target patient in each laryngoscope exposure level is counted, and preoperative airway risk indexes of the target patient can be obtained by further combining the category similarity between the indirect features and reference features. According to the method, state information in an operation is obtained in real time, state risk indexes in the operation are obtained through space state changes of the pharynx, the root of the tongue and the front wall of the trachea, the more visual EtCO2 waveform disorder degree serves as a reference, the real-time airway influence coefficient capable of representing multi-dimensional risk factors is finally obtained, the real-time airway influence coefficient is used for prediction of a classification neural network, and a better prediction result is obtained. According to the method, feature analysis is carried out on the patient group before the operation, the physiological state of the patient is analyzed during the operation, effective risk indexes are comprehensively obtained and used for difficult airway prediction, the prediction accuracy is improved, and doctors can conveniently intervene in the subsequent operation process in advance.

Description

Technical Field

[0001] This invention relates to the field of physiological data processing technology, specifically to an artificial intelligence-based system and method for predicting difficult airways. Background Technology

[0002] Difficult airway refers to a situation where intubation is difficult or unsuccessful in a target patient during endotracheal intubation or other airway management procedures due to anatomical structure, pathological condition, or other factors. Traditional preoperative airway assessment methods mainly rely on preoperative examinations combined with the doctor's clinical experience, such as assessment using single parameters like the patient's mouth opening, thyromental distance, neck mobility, and modified Mallampati classification.

[0003] Traditional assessment methods for predicting difficult airways in patients are primarily performed under static conditions. The physician's subjective judgment cannot fully and effectively account for the differences between different patient groups, such as obese or elderly patients. Furthermore, after general anesthesia, factors such as muscle relaxation, changes in body position, and drug effects dynamically alter the airway condition, potentially leading to unexpected difficult airways in patients with normal preoperative assessments during surgery. Ultimately, this results in inaccurate patient risk indicators used for prediction, leading to erroneous risk predictions. Summary of the Invention

[0004] To address the problems of existing technologies that rely heavily on physician subjectivity in predicting difficult airways, easily overlook patient group information, and fail to effectively predict difficult airways based on intraoperative status, this invention aims to provide an artificial intelligence-based difficult airway prediction system and method. The specific technical solution adopted is as follows: This invention proposes an artificial intelligence-based difficult airway prediction system, the system comprising: The historical patient data processing module is used to statistically analyze the preoperative status characteristics of each historical patient in each laryngoscopy exposure level in the historical database; in each laryngoscopy exposure level, the intersection of the preoperative status characteristics of all historical patients is used as the baseline feature; and other features that do not belong to any laryngoscopy exposure level are used as indirect features. The target patient comparison module is used to compare the type similarity between the target preoperative status characteristics of the target patient and the baseline characteristics of each laryngoscopy exposure level. It combines the frequency of the indirect characteristics involved in the target patient in each laryngoscopy exposure level to obtain the preoperative airway risk index of the target patient. The intraoperative status analysis module is used to analyze the changes in the spatial status of the pharynx, tongue base, and anterior tracheal wall of the target patient before and after anesthesia to obtain intraoperative status risk indicators. The difficult airway prediction module is used to fuse the preoperative airway risk indicators and intraoperative status risk indicators of the target patient based on the degree of disturbance of the EtCO2 waveform detected in real time during the operation, so as to obtain the real-time airway influence coefficient of the target patient; the real-time airway influence coefficient is input into a pre-trained classification neural network to output the difficult airway prediction result.

[0005] Furthermore, the method for obtaining the category similarity includes: For each laryngoscopy exposure level, the proportion of the number of preoperative features that are the same between the target preoperative features and the baseline features of the target patient is used as the category similarity.

[0006] Furthermore, the method for obtaining the preoperative airway risk indicators includes: The average type similarity of the target patient across all laryngoscopic exposure levels is used as the airway hierarchical complexity of the target patient. Based on the frequency of occurrence of the indirect features in each laryngoscopy exposure level and the magnitude of the laryngoscopy exposure level, the multi-level comprehensive abnormal impact performance of each indirect feature is obtained; the multi-level comprehensive abnormal impact performance of the indirect features included in the target patient is summed and averaged to obtain the risk impact coefficient of the indirect feature. Preoperative airway risk indicators are obtained based on the indirect characteristic risk impact coefficient and the complexity and interweaving of airway levels.

[0007] Furthermore, the method for obtaining the performance degree of the multi-level comprehensive anomaly includes: The frequency is weighted by the level of laryngoscope exposure to obtain the initial abnormality manifestation of indirect features in each level of laryngoscope exposure. The initial abnormality impact of indirect features at all laryngoscopic exposure levels is summed and then normalized to obtain the multi-level comprehensive abnormality impact.

[0008] Furthermore, the method for obtaining the intraoperative status risk indicators includes: The spatial state change characteristics of the pharynx are used as the weight of the spatial state change characteristics of the tongue root, and the negative correlation mapping result of the spatial state change characteristics of the pharynx is used as the weight of the spatial state change characteristics of the anterior tracheal wall. The spatial state change characteristics of the tongue root and the spatial state change characteristics of the anterior tracheal wall are weighted and summed to obtain the intraoperative state risk index.

[0009] Furthermore, the analytical methods for changes in the spatial state of the pharynx include: The change in pharyngeal air column width before and after anesthesia induction in the target patient is used as the numerator, and the pharyngeal air column width before anesthesia induction is used as the denominator to obtain the pharyngeal air column change amplitude coefficient of the target patient; the ratio of the pharyngeal air column change amplitude coefficient to the pharyngeal air column width after anesthesia induction is normalized to obtain the pharyngeal airway spatial tension of the target patient, and the pharyngeal airway spatial tension is used as the spatial state change characteristic of the pharynx.

[0010] Furthermore, the analytical methods for changes in the spatial state of the tongue root include: The ratio of tongue root thickness to mandibular length after anesthesia induction in the target patient was normalized to obtain the relative spatial occupancy of the tongue root as a characteristic of the spatial state change of the tongue root.

[0011] Furthermore, the analytical methods for changes in the spatial state of the anterior tracheal wall include: The absolute value of the difference in distance from the skin to the anterior tracheal wall between the induction of anesthesia in the target patient is normalized to obtain the anterior cervical soft tissue risk score, which is then used as a spatial state change characteristic of the anterior tracheal wall.

[0012] Furthermore, the method for obtaining the real-time airway influence coefficient includes: The difference signal between the latest band and the previous band in the real-time detected EtCO2 waveform is obtained; the difference signal is formed by calculating the absolute value of the difference between the signal points after aligning the two bands with the peak point as the reference point; the signal values ​​on the difference signal are accumulated and normalized to obtain the degree of disorder. The degree of disorder is used as the weight of the intraoperative status risk index, and the negative correlation mapping result of the degree of disorder is used as the weight of the preoperative airway risk index. The intraoperative status risk index and the preoperative airway risk index are weighted and summed to obtain the real-time airway influence coefficient.

[0013] This invention also proposes an artificial intelligence-based method for predicting difficult airways, the method comprising: In the historical database, the preoperative status characteristics of each historical patient in each laryngoscopy exposure level are statistically analyzed; in each laryngoscopy exposure level, the intersection of the preoperative status characteristics of all historical patients is used as the baseline feature; other features that do not belong to the baseline feature of any laryngoscopy exposure level are used as indirect features. By comparing the type similarity between the target preoperative status characteristics of the target patient and the baseline characteristics of each laryngoscopic exposure level, and combining the frequency of the indirect characteristics involved in the target patient in each laryngoscopic exposure level, the preoperative airway risk index of the target patient is obtained. By analyzing the changes in the spatial state of the pharynx, tongue base, and anterior tracheal wall before and after anesthesia in the target patients, intraoperative status risk indicators can be obtained. Based on the degree of disturbance of the EtCO2 waveform detected in real time during the operation, the preoperative airway risk indicators and intraoperative status risk indicators of the target patient are fused to obtain the real-time airway influence coefficient of the target patient; the real-time airway influence coefficient is input into a pre-trained classification neural network to output the difficult airway prediction result.

[0014] The present invention has the following beneficial effects: To analyze the relationship between patients and historical groups, this invention first divides historical groups into different laryngoscopy exposure levels. Then, within each level, common preoperative characteristics are statistically analyzed as baseline characteristics for the corresponding group, facilitating comparison between the target patient and various groups. Furthermore, considering that characteristics other than the baseline characteristics can also cause differences in potential airway risks among different patients, indirect characteristics are further analyzed. The frequency of the target patient's indirect characteristics across different laryngoscopy exposure levels is statistically analyzed, and combined with the similarity between these indirect characteristics and the baseline characteristics, a preoperative airway risk index for the target patient can be obtained. In other words, the preoperative airway risk index reflects the risk complexity by analyzing the matching of the target patient's preoperative characteristics across different groups. Intraoperative status information is further acquired in real time. By observing changes in the spatial state of the pharynx, tongue base, and anterior tracheal wall, an intraoperative status risk index is obtained, reflecting the risks to the target patient's physiological tissues caused by anesthesia. By using the more intuitive degree of EtCO2 waveform disturbance as a reference, it is determined whether to focus more on the complexity of the preoperative characteristics of the target patient or on the intraoperative physiological state after anesthesia. This ultimately yields a real-time airway influence coefficient that characterizes multidimensional risk factors, which is used for prediction by a classification neural network, resulting in better prediction results. This invention improves prediction accuracy by combining preoperative patient characteristic analysis with intraoperative analysis of patient physiological state, thus obtaining an effective risk index for predicting difficult airways and facilitating early intervention by physicians in subsequent surgical procedures. Attached Figure Description

[0015] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a block diagram of an artificial intelligence-based difficult airway prediction system provided as an embodiment of the present invention. Detailed Implementation

[0017] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of an artificial intelligence-based difficult airway prediction system and method proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0018] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0019] The following description, in conjunction with the accompanying drawings, details the specific implementation of an artificial intelligence-based difficult airway prediction system and method provided by this invention.

[0020] Please see Figure 1 The diagram illustrates a block diagram of an artificial intelligence-based difficult airway prediction system according to an embodiment of the present invention. The system includes: a historical patient data processing module 101, a target patient comparison module 102, an intraoperative status analysis module 103, and a difficult airway prediction module 104.

[0021] It should be noted that the system proposed in this embodiment of the invention relies on existing medical systems. During the preoperative and intraoperative stages, the detection results of doctors and instruments are uploaded and stored on a server. The system proposed in this embodiment of the invention obtains the final prediction result for difficult airways by calling and analyzing the data. The detection of preoperative state characteristics mainly includes anatomical structural abnormalities, case physiological factors, and other factors. Other implementations of this embodiment of the invention may select other categories of state characteristics as the patient's preoperative characteristics. Here, only the meaning and specific content of each preoperative state characteristic under one specific implementation of this embodiment of the invention are briefly described: (1) Anatomical structural abnormalities: Craniofacial deformities: such as micrognathia (receding mandible), short neck, high larynx, macroglossia, and protruding incisors. These features directly lead to a narrow space in the oral cavity, pharynx, and larynx, making it difficult to expose the glottis; Limited mouth opening: A normal mouth opening should be greater than 3 fingers (approximately 4-5 cm). Temporomandibular joint disorders, joint ankylosis, oral and maxillofacial fibrosis, or post-radiotherapy can all lead to difficulty opening the mouth. Limited neck mobility: Reduced neck flexion and extension mobility (such as in rheumatoid arthritis, ankylosing spondylitis, cervical spine injury, and morbid obesity) makes it impossible to form the standard "floral olfaction position," resulting in difficulty in aligning the glottal axis with the oral cavity axis; Morbid obesity: In addition to increased neck circumference, factors such as fat accumulation in the pharynx and thickening of the chest wall also lead to difficulties in airway management.

[0022] (2) Pathophysiological factors: Infections and inflammations, such as epiglottitis, laryngeal edema, and maxillofacial cellulitis, can lead to swelling of the airway mucosa and narrowing of the lumen. Tumors and space-occupying lesions: Tumors in the oral cavity, pharynx, neck and mediastinum can directly compress or invade the airway and change its normal anatomical structure; Trauma: Maxillofacial and neck trauma can lead to fractures, bleeding, edema, or airway rupture, causing anatomical disarray and blockage of the "oxygen delivery channel"; Other conditions: such as sleep apnea syndrome (OSA) patients are at high risk of difficult airways.

[0023] (3) Other factors: There is a history of difficult airway disease.

[0024] All of the above factors can be quantified into two types of data, 0 and 1, for recording and storage. 0 indicates that the factor is not present, and 1 indicates that the factor is present. Ultimately, a preoperative state feature sequence can be formed. In this embodiment of the invention, each patient has a preoperative state feature sequence of length 9, and the preoperative feature factors corresponding to the elements at the same position in the preoperative state feature sequences of different patients are the same.

[0025] Information collected during the surgical phase may include continuous end-tidal carbon dioxide data and spatial information data of the pharynx, base of tongue, and anterior tracheal wall obtained by ultrasound scanning of the neck before and after anesthesia induction.

[0026] Existing medical information systems, after collecting and storing data, provide a set of data for each patient. For patients who have undergone surgery, the database contains complete historical data corresponding to their status as historical patients, and doctors can classify historical patients according to the level of laryngoscopy exposure. In this embodiment of the invention, four levels of laryngoscopy exposure are set, with higher levels indicating a greater degree of difficulty in laryngoscopy exposure during the patient's historical surgery.

[0027] In order to effectively analyze the condition of the target patient before surgery, this embodiment of the invention compares the target patient with historical patients at various laryngoscopic exposure levels, and then assesses the risk characteristics of the target patient under multiple laryngoscopic exposure levels based on the preoperative condition characteristics.

[0028] To facilitate effective comparison among historical patient groups of the target patient, the historical patient data processing module 101 in the system proposed in this embodiment of the invention first performs statistical analysis on historical patients in the historical database for each laryngoscopy exposure level. Within each laryngoscopy exposure level, the intersection of the preoperative status characteristics of all historical patients is used as the baseline feature; other features not belonging to any laryngoscopy exposure level are used as indirect features. The spinal diaphragm represents the common characteristics exhibited by historical patients within the same level; while indirect features are those that do not share commonalities across all laryngoscopy exposure levels, but these features still exhibit a certain distribution across different groups, therefore the influence of indirect features at each level cannot be ignored.

[0029] The target patient comparison module 102 further compares the preoperative state feature sequence of the target patient with the baseline features of each laryngoscopy exposure level to determine the category similarity between the target patient and each level. A higher category similarity indicates that the preoperative state of the target patient more closely matches the common baseline features of the corresponding laryngoscopy exposure level group. Regarding the indirect features of the target patient, the higher the frequency of the indirect features in each group and the higher the level of the corresponding group, the greater the impact of the indirect features on the severe and difficult airway outcome. Therefore, by further combining the frequency of the indirect features involved in the target patient in each laryngoscopy exposure level, the preoperative airway risk index of the target patient is obtained. That is, the preoperative airway risk index is obtained by fusing two features: one feature characterizes the matching complexity of the target patient in multi-level groups, and the other feature characterizes the degree of influence of the preoperative state features of the target patient on the risk of severe and difficult airway.

[0030] Preferably, in this embodiment of the invention, the method for obtaining category similarity includes: For each laryngoscopy exposure level, the proportion of the number of preoperative features that are the same between the target preoperative features and the baseline features of the target patient is used as the category similarity. That is, the numerator of the category similarity is the number of target preoperative features that are the same as the baseline features of the target patient, and the denominator is the number of baseline features. The larger the ratio, the higher the category similarity and the greater the match between the target patient and the laryngoscopy exposure level.

[0031] Preferably, in this embodiment of the invention, the method for obtaining preoperative airway risk indicators includes: The average type similarity of the target patient across all laryngoscopic exposure levels is used as the airway hierarchical complexity degree of the target patient. That is, if the target patient has a strong type similarity across multiple groups, it indicates that the target patient meets the baseline characteristics of multiple groups and exhibits relatively complex physiological characteristics. In this case, the greater the corresponding average type similarity, that is, the greater the airway hierarchical complexity degree, the more complex the patient's condition and the higher the risk.

[0032] Based on the frequency of the indirect features in each laryngoscopy exposure level and the magnitude of the laryngoscopy exposure level, the multi-level comprehensive abnormal impact of each indirect feature is obtained. That is, for each indirect feature, the higher its frequency in a certain group and the higher the corresponding laryngoscopy exposure level in that group, the greater its impact on difficult airway phenomena in that group. Therefore, the multi-level comprehensive abnormal impact of each indirect feature can be obtained by combining all groups.

[0033] The risk impact coefficient of indirect characteristics is obtained by summing and averaging the multi-level comprehensive abnormality manifestations of the indirect characteristics included in the target patient.

[0034] Preoperative airway risk indicators are obtained based on the indirect characteristic risk impact coefficient and the complexity and interweaving of airway levels.

[0035] Furthermore, in this embodiment of the invention, the method for obtaining the performance degree of multi-level comprehensive anomalies includes: By using the level of laryngoscope exposure as a weight, the frequency is weighted to obtain the initial abnormality manifestation of indirect features in each level of laryngoscope exposure.

[0036] It should be noted that, in this embodiment of the invention, the method for quantifying levels into weights uses the level as the numerator and the sum of all levels as the denominator to obtain the weight of the corresponding level. The larger the result, the higher the corresponding laryngoscope exposure level, and the sum of the weights corresponding to all laryngoscope exposure levels equals a positive integer 1. By multiplying the weight by the corresponding frequency, the initial abnormality impact performance can be obtained. The initial abnormality impact performance characterizes the degree of difficulty in airway impact manifested by indirect features in a group of individual laryngoscope exposure levels.

[0037] The initial abnormal impact performance of indirect features across all laryngoscopy exposure levels is summed and then normalized to obtain the multi-level comprehensive abnormal impact performance. In this embodiment of the invention, the normalization method employs range standardization, which quantizes the data to between 0 and 1 and eliminates the influence of dimensions by statistically analyzing the maximum and minimum values ​​within this dimension.

[0038] It should be noted that the normalization proposed in the subsequent process of the embodiments of the present invention can all be implemented using the same method. Both the range standardization and the linear function mapping method are well known to those skilled in the art, and will not be described in detail here.

[0039] It should be noted that, because in one specific implementation of this invention, the multi-level comprehensive abnormality impact performance is a normalized result, the calculated indirect characteristic risk impact coefficient is also a normalized result. Therefore, the indirect characteristic risk impact coefficient can be used as a gain term and fused with the airway level complexity overlap. In this embodiment of the invention, the positive integer 1 is added to the indirect characteristic risk impact coefficient to obtain the gain term, and the product of the gain term and the airway level complexity overlap is used as the preoperative airway risk indicator.

[0040] For the target patient, the use of general anesthesia induction drugs and muscle relaxants leads to a loss of muscle tone (extreme relaxation of all muscles in the mandible, tongue, larynx, and epiglottis). Furthermore, gravity can alter the shape and diameter of the airway, potentially causing a seemingly unobstructed airway to narrow after induction, increasing ventilation resistance and intubation difficulty. Therefore, the system proposed in this embodiment further analyzes the intraoperative characteristics of the target patient. The intraoperative state analysis module 103 retrieves real-time physiological structural data of the target patient before and after anesthesia, including spatial data of the pharynx, tongue base, and anterior tracheal wall. It analyzes the spatial state changes of these physiological structures before and after anesthesia induction to obtain intraoperative state risk indicators. That is, the more dangerous the spatial state changes, the greater the risk of creating a difficult airway during subsequent surgery. The pharynx is the most prone to collapse in the upper respiratory tract. After anesthesia induction, muscle relaxation and gravity cause the base of the tongue and the posterior pharyngeal wall to move inward, directly compressing this space. Thus, the width of the pharyngeal air column directly reflects the final effective space of the airway. The tongue is the organ that occupies the largest space in the oropharynx. Its thickness determines the potential displacement and volume of the tongue falling backward after the loss of muscle tension caused by anesthesia induction. A thickened tongue base is a major hidden danger for airway obstruction. The space of the anterior tracheal wall refers to the distance from the skin to the anterior tracheal wall. An increase in this distance usually indicates the abundance of soft tissue (fat, edema) in the anterior neck. The weight of abundant tissue may exert external pressure on the trachea when the patient is supine. Furthermore, the increase in this distance also means that laryngeal anatomical landmarks (such as the thyroid cartilage and cricoid cartilage) are more difficult to palpate and locate. This is a fatal obstacle when an emergency cricothyroidotomy is required. In other words, when emergency ventilation is required through an incision, an increased distance from the skin to the anterior trachea is a risky hazard.

[0041] Preferably, in this embodiment of the invention, considering that the greater the tension in the patient's pharyngeal airway space, the more emphasis can be placed on the relative space occupancy of the patient's tongue root, thereby reflecting the direct risk brought about by the degree to which the tongue root occupies the airway space due to the influence of anesthesia induction within a limited space; conversely, the smaller the tension in the pharyngeal airway space, the more emphasis can be placed on the risk hazard of the patient's anterior neck soft tissue, that is, the spatial state change characteristics of the anterior tracheal wall, thereby reflecting the indirect risk hazard that may be caused by the need for emergency ventilation in an emergency. Based on this, the method for obtaining intraoperative status risk indicators in this embodiment of the invention includes: The spatial state change characteristics of the pharynx are used as the weight of the spatial state change characteristics of the tongue root, and the negative correlation mapping result of the spatial state change characteristics of the pharynx is used as the weight of the spatial state change characteristics of the anterior tracheal wall. That is, the weights of the spatial state change characteristics of the tongue root and the anterior tracheal wall are negatively correlated, and this weight is reflected by the spatial state change characteristics of the pharynx.

[0042] The intraoperative status risk index is obtained by weighted summing of the spatial state change characteristics of the tongue root and the spatial state change characteristics of the anterior trachea.

[0043] Furthermore, in order to quantify the spatial changes of various physiological structures before and after anesthesia induction, the method for analyzing the spatial state changes of the pharynx in this embodiment of the invention includes: This invention focuses on both the width of the induced posterior pharyngeal air column and the range of its width variation. If the width of the induced posterior pharyngeal air column is smaller and the range of variation is larger, it indicates that the final effective space of the ventilation channel is smaller.

[0044] Therefore, the change in pharyngeal air column width before and after anesthesia induction in the target patient is used as the numerator, and the pharyngeal air column width before anesthesia induction is used as the denominator to obtain the pharyngeal air column change amplitude coefficient for the target patient. The change in pharyngeal air column width is the difference between the pharyngeal air column width before anesthesia induction and the pharyngeal air column width after anesthesia induction. Therefore, the larger the pharyngeal air column change amplitude coefficient, the greater the reduction in pharyngeal air column width after anesthesia induction in the target patient, and the worse the ventilation status of the target patient.

[0045] The ratio of the pharyngeal air column variation amplitude coefficient to the pharyngeal air column width after anesthesia induction is normalized to obtain the pharyngeal airway spatial tension of the target patient. The purpose of using the pharyngeal air column width after anesthesia induction as the denominator is to establish a negative correlation mapping relationship: the smaller the pharyngeal air column width after anesthesia induction, the worse the patient's ventilation status, resulting in a smaller denominator and a larger pharyngeal airway spatial tension. Furthermore, a larger numerator indicates a greater variation amplitude caused by anesthesia, which also represents a worse pharyngeal ventilation status in the target patient, i.e., a larger pharyngeal airway spatial tension. This pharyngeal airway spatial tension is used as a characteristic of the spatial state change of the pharynx.

[0046] Furthermore, in this embodiment of the invention, considering that merely considering the change in tongue root thickness as a change in spatial state is insufficient, and that the corresponding mandibular space capacity should also be considered, the method for analyzing the change in spatial state of the tongue root in this embodiment of the invention includes: The ratio of tongue root thickness to mandibular length after anesthesia induction in the target patient was normalized to obtain the relative space occupancy of the tongue root as a characteristic of the spatial state change of the tongue root. That is, the smaller the mandibular length, the smaller the space formed by the mandible; at the same time, the greater the tongue root thickness, the greater the space occupied, and the greater the relative space occupancy of the tongue root.

[0047] Furthermore, in this embodiment of the invention, the method for analyzing the spatial state changes of the anterior tracheal wall includes: The absolute value of the difference in distance from the skin to the anterior tracheal wall between the induction of anesthesia in the target patient is normalized to obtain the anterior cervical soft tissue risk score, which is then used as a spatial state change characteristic of the anterior tracheal wall.

[0048] It should be noted that the above normalization methods can all be implemented using existing normalization methods such as range standardization. The specific details have been described in the above embodiments and will not be repeated here.

[0049] It should be noted that since all obtained spatial state change features are normalized results, the quantification of intraoperative status risk indicators can be achieved by directly multiplying the spatial state change features of the pharynx with those of the tongue root to weight the tongue root spatial state change features; subtracting the spatial state change features of the pharynx from the positive integer 1 to achieve a negative correlation mapping; and multiplying the negative correlation mapping result with the spatial state change features of the anterior tracheal wall to weight the anterior tracheal wall spatial state change features. Finally, the sum of the two weighted results is used as the intraoperative status risk indicator.

[0050] The system proposed in this embodiment of the invention further considers the need for real-time monitoring of the patient's EtCO2 waveform during the entire anesthesia process to promptly detect abnormal ventilation. The regularity of the EtCO2 waveform reflects the effective ventilation performance during the operation. When the patient's EtCO2 waveform is consistently regular, the patient's ventilation is usually good; however, when the waveform is disordered, it indicates that the patient is affected by tracheal insertion or ventilation obstruction, which may lead to a difficult airway. Therefore, the difficult airway prediction module 104 uses the degree of disorder of the EtCO2 waveform as a reference to determine whether to focus more on the complexity of the preoperative characteristics of the target patient or on the intraoperative physiological state after anesthesia. Ultimately, it obtains a real-time airway influence coefficient that characterizes multidimensional risk factors, which is used for prediction by a classification neural network. The real-time airway influence coefficient is input into a pre-trained classification neural network, which outputs the difficult airway prediction result. Finally, by performing preoperative characteristic analysis of the patient group and intraoperative physiological state analysis, an effective risk index is obtained for difficult airway prediction, improving prediction accuracy and facilitating early intervention by physicians in subsequent surgical procedures.

[0051] In this embodiment of the invention, a fully connected neural network can be selected as the classification neural network. This embodiment uses case data from over 30,000 historical patients as the training dataset. The real-time airway impact coefficients and real-time EtCO2 waveforms corresponding to the training dataset are used as input data, and manually labeled difficult airway warning values ​​are used as output data. The training set and validation set are divided in a 7:3 ratio. The neural network is trained using the training set samples, with the cross-entropy function as the loss function. Gradient descent is used to train until the damage function converges. The robustness of the training results is verified using the validation set, thus obtaining the trained neural network. The specific structure and training method of the neural network are well-known techniques to those skilled in the art, and will not be elaborated upon in this embodiment.

[0052] Preferably, in this embodiment of the invention, the method for obtaining the real-time airway influence coefficient includes: The difference signal between the latest band and the previous band in the real-time detected EtCO2 waveform is acquired. This difference signal is formed by calculating the absolute value of the difference between signal points after aligning the two bands with their peak points as reference points. The signal values ​​on the difference signal are accumulated and normalized to obtain the degree of disorder. It should be noted that in this embodiment, a band refers to a signal segment composed of a minimum value, a maximum value, and a minimum value. Each band in the signal can be obtained through an extreme value detection method, a technique well-known to those skilled in the art, and will not be elaborated upon here.

[0053] The degree of disorder is used as the weight of the intraoperative status risk index, and the negative correlation mapping result of the degree of disorder is used as the weight of the preoperative airway risk index. The intraoperative status risk index and the preoperative airway risk index are weighted and summed to obtain the real-time airway influence coefficient.

[0054] It should be noted that since the disorder level has been normalized, the negative correlation mapping result can also be obtained by subtracting the disorder level from the positive integer 1. The weighted summation method has been explained in the above embodiments and will not be repeated here.

[0055] In summary, this invention categorizes historical patient groups into different laryngoscopy exposure levels. Within each level, common preoperative characteristics are statistically analyzed as baseline features. Further analysis of indirect features reveals the frequency of these features across different laryngoscopy exposure levels. Combining these indirect features with the baseline features yields preoperative airway risk indicators for the target patient. Real-time acquisition of intraoperative status information, through spatial changes in the pharynx, tongue base, and anterior tracheal wall, provides intraoperative risk indicators. Using the more intuitive EtCO2 waveform disturbance as a reference, a real-time airway influence coefficient characterizing multidimensional risk factors is obtained, which is then used for prediction in a classification neural network, resulting in better prediction outcomes. This invention, through preoperative feature analysis of the patient group and intraoperative physiological analysis, comprehensively obtains effective risk indicators for predicting difficult airways, improving prediction accuracy and facilitating early intervention by physicians in subsequent surgical procedures.

[0056] Based on the same inventive concept, this invention also proposes an artificial intelligence-based method for predicting difficult airways, the method comprising: In the historical database, the preoperative status characteristics of each historical patient in each laryngoscopy exposure level are statistically analyzed; in each laryngoscopy exposure level, the intersection of the preoperative status characteristics of all historical patients is used as the baseline feature; other features that do not belong to the baseline feature of any laryngoscopy exposure level are used as indirect features. By comparing the type similarity between the target preoperative status characteristics of the target patient and the baseline characteristics of each laryngoscopic exposure level, and combining the frequency of the indirect characteristics involved in the target patient in each laryngoscopic exposure level, the preoperative airway risk index of the target patient is obtained. By analyzing the changes in the spatial state of the pharynx, tongue base, and anterior tracheal wall before and after anesthesia in the target patients, intraoperative status risk indicators can be obtained. Based on the degree of disturbance of the EtCO2 waveform detected in real time during the operation, the preoperative airway risk indicators and intraoperative status risk indicators of the target patient are fused to obtain the real-time airway influence coefficient of the target patient; the real-time airway influence coefficient is input into a pre-trained classification neural network to output the difficult airway prediction result.

[0057] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0058] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A difficult airway prediction system based on artificial intelligence, characterized in that, The system includes: The historical patient data processing module is used to statistically analyze the preoperative status characteristics of each historical patient in each laryngoscopy exposure level in the historical database; in each laryngoscopy exposure level, the intersection of the preoperative status characteristics of all historical patients is used as the baseline feature; and other features that do not belong to any laryngoscopy exposure level are used as indirect features. The target patient comparison module is used to compare the type similarity between the target preoperative status characteristics of the target patient and the baseline characteristics of each laryngoscopy exposure level. It combines the frequency of the indirect characteristics involved in the target patient in each laryngoscopy exposure level to obtain the preoperative airway risk index of the target patient. The intraoperative status analysis module is used to analyze the changes in the spatial status of the pharynx, tongue base, and anterior tracheal wall of the target patient before and after anesthesia to obtain intraoperative status risk indicators. The difficult airway prediction module is used to fuse the preoperative airway risk indicators and intraoperative status risk indicators of the target patient based on the degree of disturbance of the EtCO2 waveform detected in real time during the operation, so as to obtain the real-time airway influence coefficient of the target patient; the real-time airway influence coefficient is input into a pre-trained classification neural network to output the difficult airway prediction result.

2. The difficult airway prediction system based on artificial intelligence according to claim 1, characterized in that, The methods for obtaining the category similarity include: For each laryngoscopy exposure level, the proportion of the number of preoperative features that are the same between the target preoperative features and the baseline features of the target patient is used as the category similarity.

3. The difficult airway prediction system based on artificial intelligence according to claim 1, characterized in that, The methods for obtaining the preoperative airway risk indicators include: The average type similarity of the target patient across all laryngoscopic exposure levels is used as the airway hierarchical complexity of the target patient. Based on the frequency of occurrence of the indirect features in each laryngoscopy exposure level and the magnitude of the laryngoscopy exposure level, the multi-level comprehensive abnormal impact performance of each indirect feature is obtained; the multi-level comprehensive abnormal impact performance of the indirect features included in the target patient is summed and averaged to obtain the risk impact coefficient of the indirect feature. Preoperative airway risk indicators are obtained based on the indirect characteristic risk impact coefficient and the complexity and interweaving of airway levels.

4. The difficult airway prediction system based on artificial intelligence according to claim 3, characterized in that, The method for obtaining the performance degree of the multi-level comprehensive anomaly includes: The frequency is weighted by the level of laryngoscope exposure to obtain the initial abnormality manifestation of indirect features in each level of laryngoscope exposure. The initial abnormality impact of indirect features at all laryngoscopic exposure levels is summed and then normalized to obtain the multi-level comprehensive abnormality impact.

5. The difficult airway prediction system based on artificial intelligence according to claim 1, characterized in that, The methods for obtaining the intraoperative status risk indicators include: The spatial state change characteristics of the pharynx are used as the weight of the spatial state change characteristics of the tongue root, and the negative correlation mapping result of the spatial state change characteristics of the pharynx is used as the weight of the spatial state change characteristics of the anterior tracheal wall. The spatial state change characteristics of the tongue root and the spatial state change characteristics of the anterior tracheal wall are weighted and summed to obtain the intraoperative state risk index.

6. The difficult airway prediction system based on artificial intelligence according to claim 5, characterized in that, Methods for analyzing changes in the spatial state of the pharynx include: The change in pharyngeal air column width before and after anesthesia induction in the target patient is used as the numerator, and the pharyngeal air column width before anesthesia induction is used as the denominator to obtain the pharyngeal air column change amplitude coefficient of the target patient; the ratio of the pharyngeal air column change amplitude coefficient to the pharyngeal air column width after anesthesia induction is normalized to obtain the pharyngeal airway spatial tension of the target patient, and the pharyngeal airway spatial tension is used as the spatial state change characteristic of the pharynx.

7. The difficult airway prediction system based on artificial intelligence according to claim 5, characterized in that, Methods for analyzing changes in the spatial state of the tongue root include: The ratio of tongue root thickness to mandibular length after anesthesia induction in the target patient was normalized to obtain the relative spatial occupancy of the tongue root as a characteristic of the spatial state change of the tongue root.

8. The difficult airway prediction system based on artificial intelligence according to claim 5, characterized in that, Methods for analyzing changes in the spatial state of the anterior tracheal wall include: The absolute value of the difference in distance from the skin to the anterior tracheal wall between the induction of anesthesia in the target patient is normalized to obtain the anterior cervical soft tissue risk score, which is then used as a spatial state change characteristic of the anterior tracheal wall.

9. The difficult airway prediction system based on artificial intelligence according to claim 1, characterized in that, The method for obtaining the real-time airway influence coefficient includes: The difference signal between the latest band and the previous band in the real-time detected EtCO2 waveform is obtained; the difference signal is formed by calculating the absolute value of the difference between the signal points after aligning the two bands with the peak point as the reference point; the signal values ​​on the difference signal are accumulated and normalized to obtain the degree of disorder. The degree of disorder is used as the weight of the intraoperative status risk index, and the negative correlation mapping result of the degree of disorder is used as the weight of the preoperative airway risk index. The intraoperative status risk index and the preoperative airway risk index are weighted and summed to obtain the real-time airway influence coefficient.

10. A method for predicting difficult airways based on artificial intelligence, characterized in that, The method includes: In the historical database, the preoperative status characteristics of each historical patient in each laryngoscopy exposure level are statistically analyzed; in each laryngoscopy exposure level, the intersection of the preoperative status characteristics of all historical patients is used as the baseline feature; other features that do not belong to the baseline feature of any laryngoscopy exposure level are used as indirect features. By comparing the type similarity between the target preoperative status characteristics of the target patient and the baseline characteristics of each laryngoscopic exposure level, and combining the frequency of the indirect characteristics involved in the target patient in each laryngoscopic exposure level, the preoperative airway risk index of the target patient is obtained. By analyzing the changes in the spatial state of the pharynx, tongue base, and anterior tracheal wall before and after anesthesia in the target patients, intraoperative status risk indicators can be obtained. Based on the degree of disturbance of the EtCO2 waveform detected in real time during the operation, the preoperative airway risk indicators and intraoperative status risk indicators of the target patient are fused to obtain the real-time airway influence coefficient of the target patient; the real-time airway influence coefficient is input into a pre-trained classification neural network to output the difficult airway prediction result.

Images