Artificial intelligence-based airway intubation image recognition method for anesthesiology department

Abstract

The invention relates to the technical field of image processing, in particular to an anesthesiology department airway intubation image recognition method based on artificial intelligence. The method comprises the following steps: acquiring an airway CT image of a patient at double view angles, constructing a multi-curvature airway model and generating a corresponding image; correcting the model image in combination with the inherent distortion parameter of the lens to obtain a distortion corrected image; secondly, the influence of different curvatures on distortion is quantified under a multi-curvature distortion correction image, and a curvature-direction-distortion correction factor mapping relation is generated; acquiring a current airway image in real-time intubation, extracting curvature features by performing pixel-level matching with the distortion correction image, associating the curvature features with the mapping relation, and determining a distortion correction index corresponding to the current airway image; and finally, performing double correction on the current airway image by combining the lens distortion parameter and the distortion correction index, and generating a high-credibility airway correction image for subsequent image recognition.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing, in particular to an anesthesia airway intubation image recognition method based on artificial intelligence. BACKGROUND

[0002] In the clinical practice of anesthesiology, airway intubation is the core operation to ensure the smooth breathing of patients, and its success rate is directly related to patient safety. In recent years, the integration of artificial intelligence and medical imaging has provided a new idea to solve this problem. By accurately identifying the airway image during airway intubation to plan the optimal intubation path, the catheter tip can be dynamically guided to avoid high-risk areas. This intelligent auxiliary system not only reduces the operation threshold, but also saves valuable time window for emergency rescue.

[0003] Current airway intubation mainly relies on endoscopic video guidance, and doctors judge the catheter position by visual observation. However, due to the design of the optical lens, it usually produces distortion, resulting in distorted airway images. The existing technology usually uses image distortion correction technology to correct the inherent distortion of the lens, but the traditional distortion model usually assumes that the imaging environment is a straight pipe, while the patient's airway usually has different degrees of curvature. The curved section also causes serious image stretching / compression distortion. Therefore, if only the inherent distortion of the lens is corrected and the individual airway curvature characteristics of the patient are ignored, the accuracy and reliability of the final corrected airway image will be low, affecting the subsequent recognition effect. SUMMARY

[0004] In order to solve the technical problem that if only the inherent distortion of the lens is corrected and the individual airway curvature characteristics of the patient are ignored, the accuracy and reliability of the final corrected airway image will be low, affecting the subsequent recognition effect, the purpose of the present application is to provide an anesthesia airway intubation image recognition method based on artificial intelligence, and the technical solution adopted is as follows: Obtain the airway CT image of the patient under double-view, obtain the airway model image of different curvatures, and obtain the airway image at the current time during airway intubation; Based on the feature points in each airway model image, correct the airway model image using the inherent distortion correction parameters of the lens, thereby obtaining the distortion correction image corresponding to each airway model image; compare the distortion correction images under different curvatures, thereby obtaining the distortion correction factors of the distortion correction image under each curvature in each preset direction; In each airway CT image, an airway wall edge line is extracted and a curvature value of each edge pixel point on the edge line is calculated; pixel points in the airway image are matched with the edge pixel points on the airway wall edge line to obtain matched edge points; based on the curvature values of the matched edge points, position conditions and distortion correction factors of a distortion correction image in each preset direction under each curvature, a distortion correction index of the airway image is determined; The airway image is subjected to distortion correction based on the distortion correction index of the airway image and the lens intrinsic distortion correction parameters, so that an airway correction image is obtained for image recognition.

[0005] Further, the method for obtaining the distortion correction image comprises: According to the lens intrinsic distortion correction parameters and the feature points in each airway model image, each airway model image is subjected to image correction by using a Brown-Conrady distortion model, so that a distortion correction image corresponding to each airway model image is obtained.

[0006] Further, the method for obtaining the distortion correction factor comprises: Based on a center point and a diagonal line of each airway model image, the airway model image is regionally divided as a region block in each preset direction; The distortion correction image with a curvature value of 0 is taken as a standard image, and other distortion correction images except the standard image are taken as to-be-analyzed images; In the same preset direction, in the region blocks of the standard image and each to-be-analyzed image, according to the feature points and the Brown-Conrady distortion model, a distortion correction factor corresponding to the region block of each to-be-analyzed image in each preset direction is obtained.

[0007] Further, the method for obtaining the airway wall edge line comprises: Each airway CT image is taken as an input of a pre-trained neural network, so that an airway region in each airway CT image is output; Based on a Canny operator, two airway wall edge lines of the airway region in each airway CT image are obtained.

[0008] Further, the method for obtaining the matched edge point comprises: In any one airway CT image, a minimum distance of each pixel point in the airway to the airway wall edge line is obtained and a distance field is generated, a maximum value point of the distance is obtained in the distance field, and a candidate point is connected by a minimum spanning tree algorithm, so that an airway center line is obtained; By using a DTW algorithm, a point on the airway center line is matched with an edge pixel point on the airway wall edge line, so that a corresponding point of each point on the airway center line on each airway wall edge line is obtained; The corresponding point of the point of the current time's intubation progress on the point of the airway center line is taken as a matching edge point.

[0009] Further, the method for obtaining the distortion correction index comprises: The distortion includes radial distortion and tangential distortion. Under each distortion, a distortion correction factor corresponding to each matching edge point is determined according to the curvature value of the matching edge point and the distortion correction factor of the distortion correction image in each preset direction under each curvature. A preset direction corresponding to each matching edge point is determined according to the viewing angle of the airway CT image to which each matching edge point belongs. A curvature vector of each matching edge point is constructed based on the preset direction and the curvature value corresponding to each matching edge point, and the curvature vectors of all the matching edge points are summed to obtain a comprehensive curvature vector corresponding to the airway image at the current time. An included angle between the comprehensive curvature vector and each preset direction is taken as a deviation angle value corresponding to each preset direction. Under each preset direction, the distortion correction factor corresponding to each matching edge point under each distortion is weighted and fused with the deviation angle value, so as to obtain the distortion correction index of the airway image at the current time under each distortion.

[0010] Further, the method for determining the distortion correction factor corresponding to each matching edge point comprises: Under each distortion, the distortion correction factors corresponding to each curvature under each preset direction are curve-fitted, so as to obtain a distortion correction factor curve of each distortion in each preset direction. The distortion correction factor corresponding to each matching edge point is obtained in the corresponding distortion correction factor curve based on the curvature value of each matching edge point.

[0011] Further, the method for obtaining the airway correction image comprises: The distortion correction index of the airway image is substituted into a Brown-Conrady distortion model, so as to correct the airway image and obtain a primary correction image. The lens intrinsic distortion correction parameter is substituted into the Brown-Conrady distortion model, so as to perform secondary correction on the primary correction image and obtain the airway correction image.

[0012] Further, the method for obtaining the feature point comprises: An angle point in each airway model image is obtained by angle point detection and taken as a feature point.

[0013] Further, the neural network can adopt a CNN.

[0014] The present application has the following beneficial effects: The airway CT image of the patient is acquired under the double view angle, and the airway model of different curvatures is constructed and the airway model image is acquired, and the airway image at the current moment in the airway intubation process is acquired. After acquiring the feature points in each airway model image, the airway model image is corrected in combination with the lens inherent distortion correction parameters, the initial interference of the lens inherent distortion is eliminated, and the "unbiased" distortion correction image is provided for subsequent curvature adaptation. Under the distortion correction image corresponding to the airway model of multiple curvatures, the influence of different bending degrees on the image distortion is quantified, the distortion correction factors in different preset directions are generated, and the precise mapping of "curvature-direction-distortion correction factor" is realized. Further, the morphological characteristics of the airway in the airway CT image of the patient are analyzed, the key geometric parameters (curvatures of the edge pixel points on the airway wall edge) are provided for individual correction, and the adaptation error caused by the dependence on the fixed model is avoided. The association between the current position in the real-time process and the morphological characteristics of the airway is established through pixel-level matching, the precise correspondence between the subsequent correction parameters and the current intubation progress is ensured, and the reliability of dynamic adaptation is improved. Then, the individual airway curvature characteristics and the distortion correction factors are fused to generate the distortion correction index of the real-time airway image. Finally, the high-credibility airway correction image at the current moment in the airway intubation process is generated through double correction (lens inherent distortion correction parameters + distortion correction index), reliable data basis is provided for subsequent intubation path planning, narrow positioning and other identification tasks, and the misjudgment risk is reduced. BRIEF DESCRIPTION OF DRAWINGS

[0015] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art and the advantages thereof, a brief introduction will be given to the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and those skilled in the art can obtain other drawings according to these drawings without any creative effort.

[0016] Figure 1 A method flowchart of an anesthesia airway intubation image recognition method based on artificial intelligence provided by an embodiment of the present application; Figure 2 A schematic diagram of an airway model under different curvatures provided by an embodiment of the present application; Figure 3 A schematic diagram of a region block in four preset directions provided by an embodiment of the present application. DETAILED DESCRIPTION

[0017] In order to further illustrate the technical means and effects taken by the present application to achieve the predetermined object of the application, the specific implementation, structure, features and effects of the artificial intelligence-based anesthesiology airway intubation image recognition method according to the present application are described in detail as follows in combination with the drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. In addition, the 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 the present application belongs.

[0019] The specific scheme of the artificial intelligence-based anesthesiology airway intubation image recognition method provided by the present application is described in detail below in combination with the drawings.

[0020] Please refer to Figure 1 which shows a method flowchart of the artificial intelligence-based anesthesiology airway intubation image recognition method provided by one embodiment of the present application, which includes the following steps: Step S1: Obtain the airway CT image of the patient under double-view angles, obtain the airway model image with different curvatures, and obtain the airway image at the current time during the airway intubation process.

[0021] Since the different curvatures of the airway wall of the patient individual will cause different effects on the distortion of the image obtained during the airway intubation process, the airway CT image of the patient is first obtained in order to obtain the curvatures at different positions of the airway wall in the subsequent process.

[0022] Specifically, before the airway intubation, the CT imaging device is used to collect the CT images of the patient under double-view angles parallel to the front and side of the face, thereby obtaining the airway CT image of the patient under double-view angles.

[0023] The magnetic sensor is installed on the medical endoscope, the current position of the endoscope in the airway is obtained in real time through the electromagnetic positioning system, and the airway image at the current time during the airway intubation process is obtained through the medical endoscope.

[0024] In the process of performing airway intubation operation, medical endoscopes are usually equipped with short focal length wide-angle lenses to expand the field of view. The physical characteristics of the wide-angle lens can cause imaging distortion, and in addition, due to the different curvatures at different positions in the airway, different curvatures can cause changes in the position and angle of the airway and the lens, thereby causing curvature distortion of the image. The specific principle is as follows: on a flat surface, the light follows the rule that the angle of incidence is equal to the angle of reflection, the direction of reflected light is predictable, and the camera captures the original shape. On a curved surface, the tangent direction at each point is different, the light is reflected / refracted at different angles at different positions, the light path diverges or converges, the incident angle is not equal to the reflection angle at this time, the curved plane causes the edge image point to deviate from the optical axis direction, and the actual imaging position deviates from the ideal plane. At this time, the magnification of the distortion is not uniform, and the distortion is more significant at the deviation position, and the distortion is geometrically magnified. And when the curvature is large, the tissue is more abrupt, causing the light incidence angle difference to increase, thereby causing the image to be more distorted (for example, the edge is stretched, the scale is distorted). On the contrary, when the curvature is small, the tissue tends to be flat, the light is more uniform, and the distortion degree is correspondingly reduced. Therefore, the greater the curvature of the airway, the greater the degree of image distortion in the image collected by the imaging device.

[0025] In order to better measure the distortion effect of different curvatures on image imaging. The distortion of the imaging device under different curvatures can be measured by constructing a model and experiment, so as to obtain the mapping relationship between the curvature and the image distortion or correction parameter. Therefore, in the embodiment of the present application, a blood vessel model is calibrated with different curvatures, the curvatures cover the curvature range of the normal human airway ( ) and 0 curvature, and the step is , thereby obtaining nine groups of airway models under different curvatures for subsequent quantification of the influence of image distortion. Please refer to Figure 2 , which shows the airway model under different curvatures in an embodiment of the present application. The width of the blood vessel model is set to 15mm, and the thickness is 3mm, which is consistent with the average size of the human airway. The blood vessel calibration model is 3D printed with a material similar to human tracheal tissue, and the inner surface of the blood vessel model is sprayed with a matte white paint. A chessboard grid with a length and width of is drawn, and the grid intersection points are used as calibration points. Then, the short focal length wide-angle lens usually equipped with a medical endoscope is used to obtain the airway model image, wherein the shooting angle should be consistent with the angle of the airway image in the airway intubation process, and in order to ensure that the curvature direction of the airway model image is the same, the lens direction should be adjusted when shooting the airway model image to obtain the airway model image under the same curvature direction.

[0026] In the embodiment of the present application, the collection and acquisition of personal information are all authorized by the relevant user, the process does not violate relevant laws and regulations, and does not violate public order and good customs.

[0027] Step S2: based on the feature points in each airway model image, rectify the airway model image by using the lens inherent distortion correction parameter, so as to obtain the distortion correction image corresponding to each airway model image; compare the distortion correction images under different curvatures, so as to obtain the distortion correction factor of the distortion correction image under each curvature in each preset direction.

[0028] Since the airway model image contains the inherent distortion of the lens, in order to avoid the inherent distortion from being superimposed during the subsequent multi-curvature correction fusion, the influence of the inherent distortion of the lens needs to be eliminated at this time. In the embodiment of the application, the airway model image is mainly rectified by using the inherent distortion correction parameter of the lens and the feature points in the airway model image, so as to obtain the distortion correction image corresponding to each airway model image.

[0029] Firstly, the imaging device (specifically refers to a medical endoscope usually equipped with a short focal length wide-angle lens in the application) is used to analyze the standard calibration board (such as a chessboard), so as to obtain the lens inherent distortion correction parameter of the imaging device, including: the imaging device is arranged at a specific distance (3cm) from the standard calibration board under a vertical angle, and an image is collected to obtain a distortion image under the angle ; under the same conditions, a standard image of the calibration board is collected by using a device without lens distortion ; for reflecting the distortion effect of the actual imaging device; provide a distortion-free reference for quantifying the distortion degree of the lens.

[0030] the corner points in and are obtained by corner point detection, the position difference of the corner points in the two images is compared, the offset of the coordinates caused by the distortion can be quantified, and the corner point coordinates in are inversely calculated by the distortion correction formula of the Brown-Conrady distortion model, so as to obtain the pixel coordinates without distortion.

[0031] The Brown-Conrady distortion model is a classical distortion correction model in the field of optical imaging, which is mainly used to solve the image deformation problem caused by the physical defects of the camera lens. This model divides the distortion into two parts, including the radial distortion component caused by the curvature of the lens, and the tangential component caused by the non-parallelism of the lens and the imaging plane. The complete coordinate transformation formula is: ; ; In the formula, represents the coordinates of the pixel point in the distortion image; , represents the square of the distance of the pixel point to the center of the image; is a radial distortion coefficient, is a tangential distortion coefficient. represents the undistorted coordinates of the pixel point after rectification.

[0032] Then, by means of an optimization algorithm, the lens intrinsic distortion rectification parameters of the current imaging device in the Brown-Conrady model are obtained according to the principle of minimizing the distance difference between the corner points in the image and the corner points in the image , wherein is a radial distortion coefficient of the image, is a tangential distortion coefficient of the image.

[0033] It should be noted that the corner point detection algorithm and the Brown-Conrady distortion model mentioned in the embodiments of the present application are all known technologies, and will not be described here.

[0034] Further, based on the feature points in each airway model image, the lens intrinsic distortion rectification parameters obtained in the foregoing are used to rectify the airway model image, so as to obtain the distortion rectification image corresponding to each airway model image.

[0035] Preferably, in an embodiment of the present application, the method for obtaining the distortion rectification image comprises: First, the corner points in each airway model image are obtained by means of corner point detection, and the corner points are taken as the feature points.

[0036] Then, according to the lens intrinsic distortion rectification parameters (the foregoing obtained ) and the feature points in each airway model image, the Brown-Conrady distortion model is used to correct the image of each airway model image, so as to obtain the distortion rectification image corresponding to each airway model image.

[0037] At this time, the distortion rectification image corresponding to each airway model image has eliminated the interference of the lens intrinsic distortion of the imaging device, and the influence of the curvature of the airway bending on the distortion can be further quantified, so as to extract the distortion rectification factors in each direction under different curvatures, and construct the mapping relationship of “curvature-direction-distortion rectification factor”.

[0038] Preferably, in an embodiment of the present application, the method for obtaining the distortion rectification factor comprises: In the embodiments of the present application, the distortion influence can be divided into four preset directions, so as to facilitate more detailed analysis, and therefore the airway model image is divided into region blocks in the four preset directions based on the center point and the diagonal line of each airway model image, please refer to Figure 3 ​Fig. 2 shows a schematic diagram of the region blocks in four preset directions in one embodiment of the present application, wherein the four preset directions are respectively denoted as up, down, left and right.

[0039] Then, the distortion correction image with a curvature value of 0 is taken as a standard image, and other distortion correction images except the standard image are taken as to-be-analyzed images.

[0040] In the region blocks of the standard image and each to-be-analyzed image in the same preset direction, distortion correction factors corresponding to the region blocks of each to-be-analyzed image in each preset direction are obtained according to the feature points (the aforementioned corner points) and the Brown-Conrady distortion model, wherein the distortion correction factors should include five, three radial distortion and two tangential distortion, the number of which corresponds to the number of lens intrinsic distortion correction parameters.

[0041] By then, five distortion correction factors of the distortion correction image corresponding to the airway model image under each curvature in each preset direction can be obtained.

[0042] Step S3: In each airway CT image, airway wall edge lines are extracted and curvature values of each edge pixel point on the edge lines are calculated; pixel points in the airway image are matched with the edge pixel points on the airway wall edge lines to obtain matched edge points; and based on the curvature values, position conditions of the matched edge points and the distortion correction factors of the distortion correction image under each curvature in each preset direction, a distortion correction index of the airway image is determined.

[0043] In a real airway environment, the direction of airway bending is often variable, and the analysis in step S2 is only performed under a specific curvature angle, which is inevitably different from the actual curvature direction of the patient's airway. Therefore, when performing image distortion correction on the airway image at the current time in the process of intubating the patient's airway, the curvature information of the patient's individual airway needs to be obtained based on the patient's airway CT image, so as to judge the influence of the curvature of the patient's individual airway on the distortion, so as to achieve the best correction effect.

[0044] Firstly, in each airway CT image, airway wall edge lines can be extracted and curvature values of each edge pixel point on the edge lines are calculated.

[0045] Preferably, in one embodiment of the present application, the method for obtaining the airway wall edge lines comprises: Each airway CT image is taken as an input of a pre-trained neural network, so as to output the airway region in each airway CT image, wherein the neural network adopts a CNN neural network.

[0046] Two airway wall edge lines of the airway region in each airway CT image are obtained based on a Canny operator.

[0047] It should be noted that the training process of the neural network is a known technology, and the specific process is not described herein; the Canny algorithm for obtaining the edge line is also a known technology, and is not described herein.

[0048] Further, because the airway image obtained in the airway intubation process is a real-time dynamic image, and the airway CT image is a static image of the overall shape of the patient's airway, in order to ensure that the two are consistent in space and shape, so as to accurately obtain the curvature feature of the airway in the airway image at the current time, the pixel points in the airway image at the current time can be matched with the edge pixel points on the airway wall edge line to obtain matching edge points for subsequent analysis of the curvature feature in the airway image at the current time.

[0049] Preferably, in an embodiment of the present application, the method for obtaining the matching edge points comprises: In any one airway CT image, the minimum distance of each pixel point in the airway to the airway wall edge line is obtained to generate a distance field, and the maximum value points of the distance are obtained in the distance field, which correspond to the candidate points of the center line of the airway, and then the candidate points are connected by the minimum spanning tree algorithm to obtain the center line of the airway.

[0050] The points on the airway center line are matched with the edge pixel points on the airway wall edge line by using the DTW algorithm, so as to obtain the corresponding points of each point on the airway center line on each airway wall edge line, wherein if there are multiple mapping points when mapping based on the DTW algorithm, one of them can be randomly selected as the corresponding point.

[0051] Since a magnetic sensor is installed on a medical endoscope, the electromagnetic positioning system can obtain the position of the current endoscope in the airway in real time, so the position of the intubation progress at the current time on the airway center line can be determined, and therefore the corresponding points of the points of the airway image at the current time on the airway center line after DTW algorithm processing are taken as the matching edge points in the airway image at the current time.

[0052] At this time, the airway intubation progress corresponding to the airway image at the current time has a matching edge point on each airway wall edge line.

[0053] It should be noted that the DTW algorithm is a known technology, and is not described herein.

[0054] After obtaining the matching edge points corresponding to the intubation progress of the airway image at the current time, the distortion correction indicators of the airway image at the current time in each preset direction can be determined according to the curvature values and position conditions of the matching edge points and the distortion correction factors of each distortion correction image in each preset direction calculated in step S2, which are used to reflect the influence degree of the curvature feature in the airway image at the current time on the image distortion.

[0055] Preferably, in one embodiment of the present application, the method for obtaining the distortion correction index comprises: The distortion includes radial distortion and tangential distortion.

[0056] Since the curvature values of the airway model image can not completely cover every kind of curvature that the airway can have, under each distortion, the distortion correction factor corresponding to each curvature in the same preset direction is fitted by using the least square method (taking the curvature value as the horizontal coordinate and taking the distortion correction factor as the vertical coordinate), so as to obtain the distortion correction factor curve of each distortion in each preset direction. At this time, since two kinds of distortion contain three radial distortions and two tangential distortions, five distortion correction factor curves can be obtained.

[0057] Then, the corresponding distortion correction factor (vertical coordinate) is obtained in the corresponding distortion correction factor curve based on the curvature value (horizontal coordinate) of each matching edge point.

[0058] Since the airway CT image is a double-view image, the two views in this embodiment of the present application are the front and side of the parallel face, so the corresponding preset direction of each matching edge point can be determined according to the view of the airway CT image to which each matching edge point belongs. Specifically, the corresponding preset direction of the two airway wall edge lines in the airway CT image parallel to the front of the face should be left and right, and the corresponding preset direction of the two airway wall edge lines in the airway CT image parallel to the side of the face should be up and down. In view of the fact that the lens direction is adjusted to obtain the airway model image in the same bending direction, the corresponding preset direction of the two airway wall edge lines in the airway CT image can be determined according to the bending direction of the airway wall edge line. Thus, the corresponding preset direction of each matching edge point can be obtained.

[0059] Further, the curvature vector of each matching edge point is constructed based on the corresponding preset direction and the curvature value of each matching edge point, and the curvature vectors of all the matching edge points are summed to obtain the comprehensive curvature vector corresponding to the airway image at the current time.

[0060] Since the distortion correction factor is the mapping between the curvature and the preset direction, the included angle between the comprehensive curvature vector and the preset direction is obtained, and the included angle and the distortion correction factor of the curvature in the preset direction are fused, so as to obtain the distortion correction index of the airway image at the current time.

[0061] Therefore, the included angles between the comprehensive curvature vector and each preset direction are calculated and taken as the deviation angle values corresponding to each preset direction. If the deviation angle value between the comprehensive curvature vector and one of the preset directions is denoted as , then the remaining deviation angle values are , and .

[0062] Finally, in each preset direction, the distortion correction factor corresponding to the deviation angle value of the matching edge point under each distortion is fused to obtain the distortion correction index of the airway image at the current time under each distortion.

[0063] The formula model of the distortion correction index is: , wherein, represents the distortion correction index; represents the distortion correction factor of the matching edge point in the first preset direction; represents the distortion correction factor of the matching edge point in the second preset direction; represents the distortion correction factor of the matching edge point in the third preset direction; represents the distortion correction factor of the matching edge point in the fourth preset direction; represents the deviation angle value between the comprehensive curvature vector and the first preset direction.

[0064] At this point, the distortion correction index of the airway image at the current time can be obtained. Similarly, the distortion correction index should have five, including three radial distortions and two tangential distortions.

[0065] Step S4: Based on the distortion correction index of the airway image and the lens inherent distortion correction parameter, the airway image is corrected for distortion to obtain an airway corrected image for image recognition.

[0066] Based on the foregoing steps, the distortion correction index of the airway image at the current time can be obtained, which represents the influence of the airway curvature on the distortion. Therefore, in order to obtain the real image in the airway at the current time during the airway intubation process, the airway image captured by the medical endoscope can be corrected twice based on the distortion correction index of the airway image and the lens inherent distortion correction parameter to obtain an airway corrected image without image distortion.

[0067] Preferably, in an embodiment of the present application, the method for obtaining the airway corrected image comprises: The distortion correction index of the airway image is substituted into the Brown-Conrady distortion model to correct the airway image and obtain a primary corrected image. The purpose of the image distortion correction here is to eliminate the influence of the airway wall bending on the image distortion.

[0068] Then, the lens inherent distortion correction parameter is substituted into the Brown-Conrady distortion model to perform secondary correction on the primary corrected image to obtain the airway corrected image. The second image distortion correction here is to eliminate the inherent distortion of the camera.

[0069] The airway correction image at this time can more truly reflect the actual situation in the airway at the current time, so that image recognition and other processes based thereon can more accurately guide the doctor to perform airway intubation, effectively avoiding unexpected situations that may occur during airway intubation.

[0070] To sum up, the airway CT image of the patient is acquired under double-view, and the airway model of different curvatures is constructed and the airway model image is acquired, and the airway image at the current time during airway intubation is acquired. After acquiring the feature points in each airway model image, the airway model image is corrected in combination with the lens inherent distortion correction parameters, the initial interference of lens inherent distortion is eliminated, and "unbiased" distortion correction image is provided for subsequent curvature adaptation. Under the distortion correction image corresponding to the airway model of multiple curvatures, the influence of different bending degrees on image distortion is quantified, and the distortion correction factors in different preset directions are generated, realizing the accurate mapping of "curvature-direction-distortion correction factor". Further, the morphological characteristics of the airway in the airway CT image of the patient are analyzed, key geometric parameters (curvatures of edge pixel points on the airway wall edge) are provided for individual correction, and adaptation errors caused by relying on fixed models are avoided. The association between the current position in the real-time process and the morphological characteristics of the airway is established through pixel-level matching, ensuring the accurate correspondence between the subsequent correction parameters and the current intubation progress, and improving the reliability of dynamic adaptation. Then, the individual airway curvature characteristics and the distortion correction factors are fused to generate the distortion correction index of the real-time airway image. Finally, the high-credibility airway correction image at the current time during airway intubation is generated through double correction (lens inherent distortion correction parameters + distortion correction index), which provides a reliable data basis for subsequent intubation path planning, narrow positioning and other recognition tasks, and reduces the risk of misjudgment.

[0071] It should be noted that the above-mentioned sequence of the embodiments of the present application is only for description, and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multi-task processing and parallel processing are also possible or can be advantageous.

[0072] Each embodiment in the specification is described in a progressive manner, and the same and similar parts between each embodiment can be referred to each other, and each embodiment mainly describes the difference from other embodiments.

Claims

1. An artificial intelligence-based anesthesiology airway intubation image recognition method, characterized in that: The method comprises: Acquire the patient's airway CT image under dual viewing angles, obtain airway model images with different curvatures, and obtain the airway image at the current moment during airway intubation; Based on the feature points in each airway model image, the airway model image is corrected using the lens' inherent distortion correction parameters to obtain a distortion-corrected image corresponding to each airway model image. The distortion-corrected images at different curvatures are compared to obtain the distortion correction factors of the distortion-corrected image at each curvature in each preset direction. In each airway CT image, the airway wall edge line is extracted and the curvature value of each edge pixel point on the edge line is calculated; the pixel points in the airway image are matched with the edge pixel points on the airway wall edge line to obtain matching edge points; and the distortion correction index of the airway image is determined based on the curvature value and position of the matching edge points and the distortion correction factor of the distortion-corrected image in each preset direction under each curvature; The airway image is distorted and corrected based on the distortion correction index of the airway image and the inherent distortion correction parameters of the lens, thereby obtaining an airway-corrected image for image recognition.

2. The method for anesthesiology airway intubation image recognition based on artificial intelligence according to claim 1, characterized in that: The method for obtaining the distortion-corrected image includes: According to the lens inherent distortion correction parameters and the feature points in each airway model image, each airway model image is corrected using the Brown-Conrady distortion model to obtain a distortion-corrected image corresponding to each airway model image.

3. The method for anesthesiology airway intubation image recognition based on artificial intelligence according to claim 1, characterized in that: The method for obtaining the distortion correction factor includes: The airway model image is divided into regions based on the center point and diagonal line of each airway model image as region blocks in four preset directions; The distortion-corrected image with a curvature value of 0 is used as the standard image, and the other distortion-corrected images except the standard image are used as the images to be analyzed; In the same preset direction, in the area blocks of the standard image and each image to be analyzed, the distortion correction factor corresponding to the area block of each image to be analyzed in each preset direction is obtained according to the feature points and the Brown-Conrady distortion model.

4. The method for anesthesiology airway intubation image recognition based on artificial intelligence according to claim 1, characterized in that: The method for obtaining the airway wall edge line includes: Each airway CT image is used as input to a pre-trained neural network to output the airway region in each airway CT image; Based on the Canny operator, two airway wall edge lines of the airway area in each airway CT image are obtained.

5. The method for anesthesiology airway intubation image recognition based on artificial intelligence according to claim 1, characterized in that: The method for obtaining the matching edge points includes: In any airway CT image, the minimum distance between each pixel in the airway and the airway wall edge is obtained to generate a distance field. In the distance field, the maximum distance point is obtained and the candidate points are connected using a minimum spanning tree algorithm to obtain the airway centerline. Using the DTW algorithm to match the points on the airway center point with the edge pixel points on the airway wall edge line, thereby obtaining the corresponding point of each point on the airway center point on each airway wall edge line; The corresponding point of the intubation progress at the current moment on the airway centerline is used as the matching edge point.

6. The method for anesthesiology airway intubation image recognition based on artificial intelligence according to claim 1, characterized in that: The method for obtaining the distortion correction index includes: Distortion includes radial distortion and tangential distortion; Under each distortion, the distortion correction factor corresponding to each matching edge point is determined according to the curvature value of the matching edge pixel point and the distortion correction factor of the distortion-corrected image in each preset direction under each curvature; Determine a preset direction corresponding to each matching edge point according to the viewing angle of the airway CT image to which each matching edge point belongs; Constructing a curvature vector for each matching edge point based on the preset direction and curvature value corresponding to each matching edge point, and performing vector summation on the curvature vectors of all matching edge points to obtain a comprehensive curvature vector corresponding to the airway image at the current moment; The angle between the integrated curvature vector and each preset direction is used as the deviation angle value corresponding to each preset direction; In each preset direction, the distortion correction factor corresponding to each distortion of the matching edge point is weightedly fused with the deviation angle value to obtain the distortion correction index of the airway image at the current moment under each distortion.

7. The method for anesthesiology airway intubation image recognition based on artificial intelligence according to claim 6, characterized in that: Determining the distortion correction factor corresponding to each matching edge point includes: Under each distortion, the distortion correction factors corresponding to each curvature in each preset direction are curve-fitted to obtain the distortion correction factor curve of each distortion in each preset direction; The corresponding distortion correction factor is obtained in the corresponding distortion correction factor curve based on the curvature value of each matching edge point.

8. The method for anesthesiology airway intubation image recognition based on artificial intelligence according to claim 1, characterized in that: The method for acquiring the airway correction image includes: Substituting the distortion correction index of the airway image into the Brown-Conrady distortion model, thereby correcting the airway image to obtain a primary corrected image; The lens inherent distortion correction parameters are substituted into the Brown-Conrady distortion model, thereby performing a secondary correction on the primary corrected image to obtain the airway corrected image.

9. The method for anesthesiology airway intubation image recognition based on artificial intelligence according to claim 1, characterized in that: The method for obtaining the feature points includes: Corner points in each airway model image are obtained through corner detection and used as feature points.

10. The method for anesthesiology airway intubation image recognition based on artificial intelligence according to claim 4, characterized in that: The neural network may be CNN.

Images