Preparation method of head and neck model with difficult airway and difficult airway intubation training method

Abstract

The invention discloses a preparation method of a head and neck model with a difficult airway and a difficult airway intubation training method. The preparation method comprises the following steps: acquiring a first three-dimensional CT image with the difficult airway; the first three-dimensional CT image is preprocessed; marking the key anatomical region on the preprocessed first three-dimensional CT image, and generating first three-dimensional mask data according to a marking result; constructing an airway segmentation model based on a deep neural network, and performing training learning on the airway segmentation model by using the preprocessed first three-dimensional CT image and the corresponding first three-dimensional mask data to obtain a trained airway segmentation model; performing three-dimensional CT scanning on the to-be-operated patient with the difficult airway to obtain a second three-dimensional CT image; preprocessing a second three-dimensional CT image, and inputting the preprocessed second three-dimensional CT image into the trained airway segmentation model to obtain second three-dimensional mask data; and preparing a head and neck model of the patient to be operated according to the second three-dimensional mask data by using a 3D printing process.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of anesthesiology, in particular to a head and neck model preparation method with a difficult airway and a difficult airway intubation training method. BACKGROUND

[0002] Before an anesthetic operation, airway assessment of a patient is needed; preoperative airway assessment is an important means to timely find a difficult airway and reduce the occurrence of unexpected difficult airways. If the preoperative assessment result is a difficult airway, doctors usually use a head and neck model for intubation training to prevent improper handling in the operation which can lead to hypoxia, a vegetable and even death, thereby ensuring the safety of the operation.

[0003] However, the existing head and neck models mostly adopt a unified specification, and the internal structure is relatively rough, which cannot reproduce the individual characteristics of different patients; especially for special cases such as limited cervical spine mobility and nasal cavity deformity, the existing head and neck models cannot simulate the related pathological characteristics, resulting in poor preoperative intubation training effect.

[0004] The statements herein only provide background technology related to the present application, and do not necessarily constitute the prior art. SUMMARY

[0005] The present application aims to provide a head and neck model preparation method with a difficult airway and a difficult airway intubation training method, which can prepare individualized head and neck models for patients with a difficult airway to be operated on, so that medical personnel can perform targeted intubation training and reduce the risk of surgery.

[0006] In order to achieve the above-mentioned purpose, the present application realizes by the following technical scheme:

[0007] A head and neck model preparation method with a difficult airway, comprising:

[0008] Obtaining a first three-dimensional CT image with a difficult airway;

[0009] Pretreating the first three-dimensional CT image to obtain a pretreated first three-dimensional CT image;

[0010] Labeling a key anatomical region on the pretreated first three-dimensional CT image, and generating first three-dimensional mask data according to the labeling result;

[0011] Constructing an airway segmentation model based on a deep neural network, and training and learning the airway segmentation model by using the pretreated first three-dimensional CT image and the corresponding first three-dimensional mask data to obtain a trained airway segmentation model;

[0012] Performing three-dimensional CT scanning on a patient with a difficult airway to be operated on to obtain a second three-dimensional CT image;

[0013] The second three-dimensional CT image is input into the trained airway segmentation model after preprocessing to obtain second three-dimensional mask data;

[0014] According to the second three-dimensional mask data and using a 3D printing process, a head and neck model of the patient with a difficult airway is prepared.

[0015] Optionally, the preprocessing includes noise reduction processing, grayscale normalization processing, and cropping processing; and the preprocessed first three-dimensional CT image and the preprocessed second three-dimensional CT image only retain the head and neck region.

[0016] Optionally, the key anatomical regions include soft tissue and bony tissue; and the soft tissue includes a nasal cavity, an oral cavity, a throat, a laryngeal cavity, and a trachea.

[0017] The nasal cavity includes a nasal concha and a nasal septum.

[0018] The oral cavity includes a tongue, a hard palate, and a soft palate.

[0019] The throat includes a pharyngeal wall and an epiglottis.

[0020] The laryngeal cavity includes a larynx, a glottis, and a vocal cord.

[0021] The bony tissue includes a skull base, a mandible, and a cervical vertebra.

[0022] Optionally, the deep neural network is a 3D U-Net convolutional neural network or a generative adversarial network.

[0023] Optionally, the step of preparing the head and neck model of the patient with a difficult airway according to the second three-dimensional mask data and using a 3D printing process includes:

[0024] Converting the second three-dimensional mask data into an STL format file;

[0025] Using three-dimensional modeling software to divide the bony tissue in the STL format file into rigid regions and the soft tissue into flexible regions;

[0026] Using a 3D printer to print the rigid regions and the flexible regions;

[0027] Assembling the printed rigid regions and flexible regions to obtain the head and neck model.

[0028] Optionally, the nasal concha, the glottis, and the tongue in the flexible region are printed in blocks.

[0029] Optionally, the printing material of the rigid region is photosensitive resin or nylon.

[0030] Optionally, the printing material of the flexible region is an elastic material.

[0031] Optionally, the printing material of the nasal cavity and the trachea in the flexible region is transparent silicone.

[0032] In another aspect, the present application also provides a difficult airway intubation training method, comprising:

[0033] The head and neck model is prepared by the method described above;

[0034] The head and neck model is installed on an adjustable support frame;

[0035] A detachable gasket is arranged at the nasal cavity entrance and the oral cavity entrance of the head and neck model;

[0036] An intubation device is used to perform intubation training on the head and neck model.

[0037] Compared with the prior art, the present application has at least one of the following advantages:

[0038] The present application provides a head and neck model with a difficult airway and a difficult airway intubation training method, which constructs an airway segmentation model based on a deep neural network, and automatically labels key anatomical regions by using the trained airway segmentation model, so as to realize fine identification of bony tissue and soft tissue by means of deep learning, and greatly improve the labeling accuracy and identification ability compared with the simple segmentation based on two-dimensional slices or threshold method.

[0039] The present application can prepare a personalized head and neck model for a patient with a difficult airway to be operated, so that medical personnel can perform targeted intubation training and reduce the risk of operation. The present application supports personalized customization and simulation simulation for special pathological conditions such as deviation of nasal septum, tumor compression and cervical spine lesions, and has stronger pertinence for identification and processing of difficult airway, which makes up for the defect that the general head and neck model cannot meet the needs of complex cases.

[0040] In the present application, a head and neck model with a difficult airway is prepared by using a 3D printing process, and hard material, elastic material or transparent material is selected for different regions such as bony tissue and soft tissue and airway lumen for partition forming, which not only highly restores the real anatomical features in the structure level, but also is closer to the actual intubation experience in the sense of touch and mechanical feedback.

[0041] The present application constructs a complete closed-loop process from data acquisition, automatic labeling, three-dimensional mask data generation to 3D printing and clinical application, significantly reduces the risk of human operation intervention by standardized and automated means, and improves the production efficiency and accuracy of the head and neck model. BRIEF DESCRIPTION OF DRAWINGS

[0042] Figure 1 is a flow chart of a head and neck model preparation method for a difficult airway according to an embodiment of the present application;

[0043] Figure 2 is a flow chart of a difficult airway intubation training method according to an embodiment of the present application. DETAILED DESCRIPTION

[0044] The head and neck model preparation method for a difficult airway and the difficult airway intubation training method according to the present application will be further described in detail below in conjunction with the accompanying drawings and specific embodiments. The advantages and features of the present application will be more apparent according to the following description. It should be noted that the accompanying drawings are greatly simplified and all use non-precise proportions, only to facilitate, clear auxiliary purpose of explaining the embodiments of the present application. In order to make the purpose, features and advantages of the present application more apparent and easy to understand, please refer to the accompanying drawings. It should be noted that the structure, proportion, size, etc. shown in the drawings attached to the present specification are only used to cooperate with the content disclosed in the specification, so that those skilled in the art can understand and read, and are not used to limit the conditions of the implementation of the present application, so they do not have the technical meaning of substance, any modification of structure, change of proportion relationship or adjustment of size, as long as it does not affect the effect and purpose that can be achieved by the present application, it should still fall within the scope of the technical content disclosed by the present application.

[0045] As described in the background section, the inventors observed in clinical anesthesia work that the traditional head and neck model for intubation training cannot reflect the individual differences and complex pathological changes of patients due to the use of standardized design, resulting in a lack of sufficient targeted training for medical personnel during operation, increasing the clinical risk. After extensive research, the inventors found that although 3D printing technology has been applied in the field of orthopedics, dentistry, etc., there are still obvious deficiencies in reproducing the nasal cavity, throat and other soft tissues and dynamic operation scenarios.

[0046] In view of this, the present application provides a head and neck model preparation method for a difficult airway, which can reproduce the individual differences and complex pathological changes of patients, and can be used for targeted training of medical personnel. Figure 1As shown, the embodiment provides a head and neck model preparation method with a difficult airway, including: step S1, obtaining a first three-dimensional CT image with a difficult airway. Step S2, pre-processing the first three-dimensional CT image to obtain a pre-processed first three-dimensional CT image. Step S3, labeling the key anatomical region on the pre-processed first three-dimensional CT image, and generating first three-dimensional mask data according to the labeling result. Step S4, constructing an airway segmentation model based on a deep neural network, and training and learning the airway segmentation model using the pre-processed first three-dimensional CT image and the corresponding first three-dimensional mask data to obtain a trained airway segmentation model. Step S5, performing three-dimensional CT scanning on a patient with a difficult airway to be operated to obtain a second three-dimensional CT image. Step S6, pre-processing the second three-dimensional CT image and inputting it into the trained airway segmentation model to obtain second three-dimensional mask data. Step S7, preparing a head and neck model of a patient with a difficult airway to be operated according to the second three-dimensional mask data and using a 3D printing process.

[0047] Specifically, the step S1 includes: step S11, obtaining a plurality of original three-dimensional CT images from a hospital image database; step S12, screening the original three-dimensional CT images by medical image browsing software to obtain a first three-dimensional CT image.

[0048] In the step S11, a plurality of (for example, 100) three-dimensional CT images with difficult airway or suspicious difficult airway characteristics can be screened from the hospital image database and recorded as original three-dimensional CT images. The format of the original three-dimensional CT images is unified as DICOM format; the original three-dimensional CT images are scanned by high-resolution 3D CT equipment, and the scanning layer thickness is set to 1.0 mm. Optionally, the original three-dimensional CT images screened from the hospital image database should include cases of different age groups, different causes such as congenital malformation, tumor compression, cervical spine lesion, etc., so that the original three-dimensional CT images have good representativeness, thereby making the trained airway segmentation model have good applicability. Optionally, the patient's personal information in the original three-dimensional CT image is de-identified by using anonymization technology to ensure privacy security.

[0049] In the step S12, after obtaining a batch of original three-dimensional CT images, all original three-dimensional CT images are quickly screened by medical image browsing software (such as DICOM browser); if some original three-dimensional CT images have serious metal artifacts, missing key anatomical regions, or serious image distortion, etc., they are excluded, and the remaining original three-dimensional CT images are the first three-dimensional CT images. Optionally, the first three-dimensional CT images are arranged according to a predetermined file naming rule for subsequent batch operation.

[0050] Optionally, the key anatomical regions include soft tissue and bony tissue, and the soft tissue includes nasal cavity, oral cavity, throat, laryngeal cavity, trachea; wherein the nasal cavity includes turbinates and nasal septum; the oral cavity includes tongue, hard palate and soft palate; the throat includes pharyngeal wall and epiglottis; the laryngeal cavity includes larynx, glottis and vocal cords; the bony tissue includes skull base, mandible and cervical vertebrae, but the present application is not limited thereto.

[0051] Specifically, in the step S2, the preprocessing includes noise reduction processing, gray scale normalization processing and cropping processing; and the first three-dimensional CT image after preprocessing only retains the head and neck region, thereby reducing the computational load of subsequent labeling or segmentation.

[0052] More specifically, the step of preprocessing the first three-dimensional CT image includes: step S21, applying a spatial filtering method to the first three-dimensional CT image for noise reduction processing in a medical image processing software (such as 3D Slicer); optionally, the spatial filtering method selects anisotropic diffusion filtering to remove noise while retaining the edge details of the first three-dimensional CT image. Step S22, the first three-dimensional CT image after noise reduction is subjected to gray scale normalization processing, so that the contrast of soft tissue and bony tissue is clearer for subsequent computer automatic analysis and processing. Step S23, the first three-dimensional CT image after gray scale normalization processing is subjected to cropping processing to retain the head and neck region; and the first three-dimensional CT image only retaining the head and neck region is the first three-dimensional CT image after preprocessing.

[0053] Optionally, in the cropping processing, a region cropping method based on threshold and automatic ROI (Region of Interest) detection is used to limit the range of interest to the head and neck region, and retain the complete structure of key anatomical regions such as nasal cavity, throat, trachea in three-dimensional space, but the present application is not limited thereto.

[0054] Specifically, in the step S3, a medical image labeling platform (such as 3D Slicer) is used for layer-by-layer segmentation, and the key marking tool built-in the software can be used for labeling to ensure accurate outlining of the key anatomical regions. More specifically, on each first three-dimensional CT image after preprocessing, key parts such as nasal cavity (including turbinates and nasal septum), oral cavity (including tongue, hard palate and soft palate), throat (including pharyngeal wall and epiglottis), laryngeal cavity (including larynx, glottis and vocal cords), trachea and bony tissue (including skull base, mandible, cervical vertebrae, etc.) are marked to obtain corresponding labeling results. Optionally, the bony tissue is separately classified as a category to distinguish between soft and hard tissue. Optionally, the multi-label function built-in the software is used for labeling to label each key part by different colors or label names, so as to ensure that the first three-dimensional mask data generated according to the above labeling results remains consistent in structure when loaded subsequently.

[0055] In an embodiment, the above-mentioned annotation results can be cross-checked or reviewed by a clinician with rich experience or an image specialist to ensure the quality and accuracy of the above-mentioned annotation results, thereby improving the annotation (i.e. segmentation) accuracy of the subsequently trained airway segmentation model. It can be understood that if a missed or mislabeled (e.g. confusion of the nasal turbinates and nasal septum, or omission of the epiglottis edge) is found, the corresponding annotation result is immediately modified in the software. In addition, in order to reduce the phenomenon of breaking and overlapping, the entire annotation sequence is played back in the three-dimensional view mode to observe the continuity of key parts such as nasal turbinates, glottis and larynx in space; if there is obvious jump or incoherence in the key parts, it is returned to the corresponding pre-processed first three-dimensional CT image for repair, and finally the integrity and accuracy of the key parts in three-dimensional space are ensured.

[0056] Specifically, after the above-mentioned annotation results pass the review, the export function of the medical image annotation platform is used to combine or separately export the above-mentioned annotation results, i.e. all labels, into NIfTI medical image format, to generate first three-dimensional mask data that can be read by the subsequently trained airway segmentation model (AI model). Optionally, in order to improve compatibility, the first three-dimensional mask data is stored in the folder path where the pre-processed first three-dimensional CT image is located, so as to facilitate quick matching and calling in various deep learning frameworks or medical image processing tools.

[0057] Specifically, in the step S4, the pre-processed first three-dimensional CT image is loaded into the deep neural network as input data, and the corresponding first three-dimensional mask data is loaded into the deep neural network as output data, so as to train and learn the airway segmentation model constructed based on the deep neural network. Optionally, the pre-processed first three-dimensional CT image with completed annotation is divided into a training set, a validation set and a test set according to a ratio of 7:2:1; when dividing, it is ensured that airway cases of different degrees are evenly distributed in each subset, so as to avoid overfitting of the airway segmentation model to certain types of structural features due to uneven data distribution.

[0058] Optionally, according to the complexity of the head and neck region and the hardware computing power, the deep neural network selects a 3D U-Net neural network with three-dimensional convolution operation capability. When constructing the airway segmentation model, the depth of the convolution layer and the number of feature channels are moderately increased to improve the recognition accuracy in small or irregular anatomical structures. Common modules such as Batch Normalization and Residual Block are introduced into the 3D U-Net neural network to improve the convergence speed and generalization ability of the network. The PyTorch deep learning framework is used for network building and debugging, and the GPU server environment is used for training acceleration. It can be understood that in other embodiments, the deep neural network can also use a neural network with an attention mechanism, or use a generative adversarial network (GAN) to improve the segmentation accuracy of the airway edge, which is not limited here.

[0059] Optionally, when training the airway segmentation model, taking an RTX 6000 graphics card as an example, the sampling window PatchSize is set to 128x128x128, the batch size Batch Size is set to 4, and the initial learning rate is set to 2x10 -4 to meet the video memory requirements of three-dimensional convolution calculation. A composite loss function based on cross-entropy and Dice is selected to better handle the multi-class imbalance problem. After each training epoch, perform segmentation inference on the validation set, calculate the Dice coefficient, sensitivity, and specificity, and dynamically adjust the learning rate or network layer weight according to the results until the trained airway segmentation model achieves optimal convergence performance in the validation set.

[0060] Further, the trained airway segmentation model is finally evaluated on the test set, still using the Dice coefficient, sensitivity, specificity, and other indicators to test the segmentation accuracy of the trained airway segmentation model for key parts such as the turbinate, glottis, larynx, cervical spine, etc. For parts that perform poorly, check if there are inaccurate annotations or insufficient samples by backtracking the annotation results, and if necessary, moderately expand or fine-tune the data set or airway segmentation model structure to obtain a model that meets the evaluation requirements and use it as the trained airway segmentation model; at this point, the training of the airway segmentation model is completed, and the weight or parameter file of the trained airway segmentation model is saved for later calling.

[0061] In the step S5, a high-resolution 3D CT device is used to scan the patient with a difficult airway to obtain a three-dimensional CT image of the patient, which is denoted as a second three-dimensional CT image. When scanning, the patient is in a supine position with the neck moderately stretched, and the scanning layer thickness is set to 1.0 mm. It should be noted that if the initial scanning fails to clearly show the key structures such as the top of the nasopharynx, the throat or the cervical vertebrae of the patient, multiple scans can be performed as needed, and different scan sequences can be combined to better distinguish between bony tissue and soft tissue. In addition, after obtaining the second three-dimensional CT image, a medical image browsing software (such as a DICOM browser) is also used to quickly screen the second three-dimensional CT image. If the second three-dimensional CT image has serious metal artifacts, missing key anatomical regions, or serious image distortion, the patient is scanned again to ensure that the second three-dimensional CT image meets the quality requirements.

[0062] In the step S6, the preprocessing of the second three-dimensional CT image includes the following steps: in the step S61, a spatial filtering method is applied in a medical image processing software (such as 3D Slicer) to perform noise reduction processing on the second three-dimensional CT image. Optionally, the spatial filtering method selects anisotropic diffusion filtering to remove noise while preserving the edge details of the second three-dimensional CT image. In the step S62, the second three-dimensional CT image after noise reduction is subjected to gray scale normalization processing to make the contrast of soft tissue and bony tissue clearer for subsequent computer automatic analysis and processing. In the step S63, the second three-dimensional CT image after gray scale normalization processing is subjected to cropping processing to retain the head and neck region. The second three-dimensional CT image retaining only the head and neck region is the preprocessed second three-dimensional CT image. It can be understood that the preprocessed second three-dimensional CT image retains only the head and neck region, which can reduce the computational load of subsequent labeling or segmentation.

[0063] Specifically, the preprocessed second three-dimensional CT image is loaded as input data into the trained airway segmentation model to enable the trained airway segmentation model to automatically label (i.e., automatically segment) the key anatomical regions in the preprocessed second three-dimensional CT image, and generate second three-dimensional mask data according to the labeling result. The second three-dimensional mask data usually contains multiple labels such as nasal cavity, oral cavity, bony tissue, throat and trachea, etc. Optionally, the trained airway segmentation model can be deployed in a preset segmentation platform and the weight or parameter file of the model can be loaded to enable the trained airway segmentation model and the preset segmentation platform to constitute an automatic segmentation system. The automatic segmentation system automatically identifies the model class number and network depth and other information when initialized, which prepares for subsequent labeling. In addition, the preset segmentation platform has a real-time visualization function to facilitate viewing of the three-dimensional rendering effect of the labeled second three-dimensional mask data.

[0064] Further, after generating the second three-dimensional mask data, the automatic segmentation system can automatically calculate the geometric parameters (e.g. volume or area) of the corresponding parts based on the voxel count of each label in the second three-dimensional mask data, and display the key values such as the volume of the nasal cavity and the minimum cross-sectional area of the oropharynx in real time on the interface. It can be understood that the normal reference range of the volume of the nasal cavity is about 13cm 3 - 40cm 3 , and the normal reference range of the minimum cross-sectional area of the oropharynx is about 3cm 2 - 7cm 2 . If the calculated value of the volume of the nasal cavity and the calculated value of the minimum cross-sectional area of the oropharynx are lower than the corresponding lower limit, it indicates that there may be stenosis or obstruction in the cavity; if the calculated value of the volume of the nasal cavity and the calculated value of the minimum cross-sectional area of the oropharynx are higher than the corresponding upper limit, it indicates that the cavity is abnormal or over-expanded. Alternatively, medical personnel can quickly assess the airway risk of the patient to be operated and optimize the intubation plan according to the volume of the nasal cavity and the minimum cross-sectional area of the oropharynx, combined with the medical history of the patient to be operated and the endoscopic image.

[0065] Further, before performing the step S7, the second three-dimensional mask data can also be manually corrected and locally refined to improve the quality of the second three-dimensional mask data. For example, for abnormal tissues such as nasal cavity tumors, throat inflammation or cervical spine deformities that may exist in the patient to be operated, the trained airway segmentation model is prone to misclassification or missed classification when automatically segmenting; in this case, medical personnel can quickly locate these abnormal areas and evaluate the accuracy of their boundaries through three-dimensional rotation and multi-view slice browsing. For another example, if significant deviations are found locally, the edge can be modified using the built-in brush or eraser tool of the preset segmentation platform to ensure the continuity of the nasal concha, glottis, laryngeal wall and other subtle structures. The edge is fine-tuned or interpolated where it is too smooth or broken to reduce the jagged or irregular edges that occur during segmentation. Alternatively, after the correction of the second three-dimensional mask data is completed, the preset segmentation platform automatically records the modification traces for subsequent queries, but the present application is not limited thereto.

[0066] The step S7 comprises: a step S71 of converting the second three-dimensional mask data into an STL format file; a step S72 of using three-dimensional modeling software to divide the bony tissues in the STL format file into rigid regions and the soft tissues into flexible regions; a step S73 of using a 3D printer to print the rigid regions and the flexible regions; and a step S74 of assembling the printed rigid regions and flexible regions to obtain the head and neck model.

[0067] Specifically, in step S71, the second three-dimensional mask data obtained in step S6 can be converted into an STL format file for loading in a subsequent 3D printing or visualization environment; and the STL format file includes a head and neck model to be printed, which has soft tissues such as a nasal cavity, an oral cavity, a throat, a laryngeal cavity, a trachea, and bony tissues. Optionally, basic noise removal and topology repair can be performed on the STL format file using three-dimensional mesh processing software (for example, MeshLab), to ensure that the topology of the head and neck model to be printed in the STL format file is continuous and reasonable, thereby reducing the excessive density or defects of the surface patches of the head and neck model to be printed in the STL format file.

[0068] Optionally, if the head and neck model to be printed in the STL format file is too large or has tiny burrs, a mesh simplification algorithm can be used to reduce the number of surfaces while ensuring the main anatomical details, to improve the efficiency of subsequent 3D printing; and then a surface smoothing operation can be performed to eliminate local irregularities caused by manual correction or segmentation algorithms.

[0069] Optionally, the head and neck model to be printed in the STL format file is subjected to an overall test, such as verifying whether there are gaps or inverted surfaces in the head and neck model to be printed using the Volume function of three-dimensional mesh processing software, to ensure that there is no layering error in the printing process. After confirmation, the STL format file and the corresponding segmentation information are archived together, and this file can be immediately put into 3D printing to prepare a personalized head and neck model, or can be left in a teaching demonstration library for further simulation analysis.

[0070] Specifically, in step S72, the three-dimensional modeling software (for example, Geomagic) can be used to partition the head and neck model to be printed in the STL format file; the main basis for partitioning is anatomical features and model functional requirements. More specifically, the bony tissues are divided into rigid regions, and the soft tissues such as the nasal cavity, the oral cavity, the throat, the laryngeal cavity, and the trachea are divided into flexible regions, so as to match corresponding printing materials for the rigid regions and the flexible regions.

[0071] Optionally, the printing material of the rigid regions is photosensitive resin or nylon, so that the printed rigid regions have high hardness, thereby ensuring the rigidity and stability of the printed rigid regions and the head and neck model. The printing material of the flexible regions is an elastic material, so as to present a human-like touch and deformation characteristic when the flexible regions are pressed and intubated; optionally, the printing material of the nasal cavity and the trachea in the flexible regions is transparent silicone (i.e., transparent resin), so that the movement trajectory of the instrument in the trachea or the nasal cavity can be observed from the outside during intubation practice, but the present application is not limited thereto.

[0072] In addition, after the partition is completed, the partitioned STL format file can be imported into a 3D printing software (for example, GrabCAD Print) to automatically generate a support structure by the 3D printing software (for example, GrabCAD Print), the support structure being located at the connection of the bone tissue and the soft tissue, that is, the bone tissue, the soft tissue and the support structure jointly form a head and neck model to be printed; especially for the cavity or thin-walled part such as the turbinate, the laryngeal wall and the like, appropriate support is added to prevent suspension or deformation. If the automatic support is not ideal, manual support or moderate thickening can be used for key parts, so as to improve the success rate of 3D printing and ensure the integrity of the model.

[0073] Specifically, in the step S73, the STL format file after partitioning and generating a support structure can be printed by using an industrial-grade 3D printer Multi-Jet Fusion based on the selected printing material, so as to print the rigid region and the flexible region. More specifically, before the 3D printer starts to work, that is, before the printing operation is performed, parameters such as layer thickness, printing speed and resolution can be set according to the characteristics of the printing material and the specifications of the printer, so as to ensure that the rigid region and the flexible region can be successfully formed by stacking. After the 3D printer starts to work, the layer-by-layer stacking process is observed by using a built-in or externally connected monitoring system. If problems such as material blockage, local collapse or nozzle deviation occur, the printing should be paused in time and maintenance should be performed.

[0074] Optionally, considering the printing volume or reducing material consumption, the rigid region and the flexible region can be printed by using a block printing method; in this case, a mortise and tenon joint or a plug-in structure needs to be set in the design. Optionally, the turbinate, the glottis and the tongue in the flexible region are printed by using a block printing method, so that special anatomical parts such as the turbinate, the glottis and the tongue root part are detachable modules, which are convenient for individual replacement after wear or customization for different pathological conditions; in this way, the maintenance cost of the whole head and neck model can be saved, and medical staff can flexibly switch the training scene, but the present application is not limited thereto.

[0075] In other embodiments, other 3D printing technologies with multiple nozzles or partition feeding functions such as FDM composite material printing and SLS double-material composite printing can be used to print the rigid region and the flexible region respectively. The printing material can be replaced according to the cost or use frequency, such as using environmentally friendly biodegradable materials such as PLA and PCL for short-term use in teaching scenarios, or introducing metal materials to print the rigid region, as long as the soft, semi-hard and hard materials can be distinguished in the printed head and neck model or a single material is used but the cavity structure is retained.

[0076] In step S74, the printed rigid region and flexible region are assembled according to the corresponding positions, and the joint is sealed with epoxy resin or a special adhesive provided by the material manufacturer; finally, the overall shape of the head and neck model is ensured to be accurate, and the internal cavity is not obviously misaligned at the splicing position.

[0077] In addition, in some embodiments, after the printing of the head and neck model with a difficult airway is completed, quality inspection of the printed head and neck model is also required. Specifically, it includes:

[0078] (1) Precision comparison and size measurement: using a digital measurement tool (such as a three-coordinate measuring machine) to measure the size of key positions in the head and neck model, such as the intervertebral distance, nasal cavity width, oropharyngeal cross-sectional area, etc.; and comparing the measured values of the key positions in the head and neck model with the measured values of the key positions in the second three-dimensional CT image of the patient to be operated, to check whether they are within the allowable error range. Pay special attention to bone tissue and airway lumen. If the deviation is too large, the printing parameters or grid processing method need to be adjusted.

[0079] (2) Material touch and operation experience test: simple finger pressure or intubation test is performed on the tongue, pharyngeal wall, epiglottis, etc. of the flexible region to confirm the similarity of its elasticity, deformation and actual human tissue touch. If the softness is insufficient or too hard, the material formula or the thickness of the flexible material needs to be adjusted in the next printing. Use a conventional nasal catheter or laryngoscope operation to simulate intubation and judge whether the friction feeling in the narrow area of the nasal concha and nasal septum is reasonable.

[0080] (3) Overall appearance and structural stability: subjectively and objectively evaluate the overall appearance integrity and structural stability of the head and neck model, including whether there are obvious cracks, deformations or material layer separation phenomena. Local brittle parts can be tested for a limited number of bending or stretching tests to ensure that the head and neck model will not be damaged quickly in normal use environment. If the strength of the key position is insufficient, additional support or more durable material formula is added in subsequent printing.

[0081] In other embodiments, after the preparation of the head and neck model with a difficult airway is completed, preferably after the quality inspection is completed, the head and neck model can be installed on an adjustable support frame to simulate the cervical spine angle of the patient to be operated in different body positions. With the help of the adjustable support frame, medical personnel can more realistically reproduce the intubation posture in the operating room or rescue room.

[0082] Optionally, a window is reserved or a small sensor (such as a pressure sensor or a camera, etc.) is installed on the printed head and neck model to collect data or record video during the intubation process, and to observe the intubation path, force distribution, etc. in real time.

[0083] Optionally, the printed head and neck model is cleaned and dried in time to avoid the flexible material in the head and neck model from aging and deforming in a long time high temperature or humid environment. If long-term storage is required, the head and neck model is preferably placed in an incubator or light shielding box to delay the performance degradation of the material.

[0084] In other embodiments, other three-dimensional medical image data such as MRI or CBCT can be used instead of the three-dimensional CT image in the present embodiment; or when the patient has severe motion artifacts or metal implants that cause low-quality CT data, multi-modal registration technology can be introduced to fuse traditional CT data with local information of ultrasound or MRI and replace the three-dimensional CT image in the present embodiment to more finely depict the nasal cavity, oral cavity or other soft tissue details. It can be understood that the above is only a replacement of data, and the specific preparation method of the head and neck model will not change.

[0085] Based on the same inventive concept, as shown in Figure 2 The present embodiment also provides a difficult airway intubation training method, which comprises the following steps: S201, preparing the head and neck model by the method described above; S202, mounting the head and neck model on an adjustable support frame; S203, arranging detachable pads at the nasal cavity entrance and oral cavity entrance of the head and neck model; and S204, using an intubation device to perform intubation training on the head and neck model.

[0086] Specifically, in the step S202, the head and neck model that has passed quality inspection can be mounted on an adjustable support frame; the support frame can adjust the supine or lateral recumbent angle and provide fine adjustment function for the cervical spine position to simulate different clinical body position requirements. Alternatively, the head and neck model can also be mounted on a simulation device with a proportion of human torso, but the present application is not limited thereto.

[0087] Specifically, in the step S203, detachable pads are arranged at the nasal cavity entrance and oral cavity entrance of the head and neck model to buffer the frequent friction of the intubation device (including conventional laryngoscope, video laryngoscope, soft fiber bronchoscope, airway tube, etc.) on the flexible area. For the stress-prone parts of the mandible or cervical spine, a reinforcing film can also be locally attached to avoid excessive local material loss caused by long-term or repeated training.

[0088] Specifically, in the step S204, medical personnel or other operators can use various intubation devices such as conventional laryngoscope, video laryngoscope or soft fiber bronchoscope to perform actual practice on the head and neck model. The transparent silicone material is used in the head and neck model, and medical personnel can watch the view closer to the real anatomy in the display or ocular lens when using video laryngoscope or fiber scope, providing a more realistic simulation environment for clinical emergency treatment.

[0089] During intubation training, for head and neck models with special anatomical features such as nasal septum deviation, laryngeal stenosis, or tumor compression, medical personnel can use different angles and intubation paths to realistically experience the impact of narrow spaces on intubation equipment. By fine-tuning the laryngoscope or lens angle, they can understand the visual blind spots and tactile feedback differences that may occur in restricted environments, thereby accumulating practical operational experience for handling suspected difficult airway cases.

[0090] Specifically, when intubating in areas of elastic material such as the tongue root, throat wall, etc., the head and neck model will produce a certain deformation resistance, and medical personnel can experience similar pulling sensations and mechanical feedback as with real patients. If mask ventilation or oropharyngeal airway insertion is required, different angles of the simulated head and neck can also be used for process training.

[0091] Specifically, after multiple rounds of intubation or ventilation training on the head and neck model, key indicators such as the number of intubation attempts, success rate, time consumption, and the degree of wear on flexible areas in the airway are collected. Comparing these indicators with the hospital's standard airway mannequins or simple training models can help evaluate the simulation fidelity and practical value of the personalized head and neck model in difficult airway scenarios.

[0092] Specifically, if there is significant damage, delamination, or deformation at special sites such as the nasal turbinates or glottis, it indicates that material selection or partition design still needs improvement. In the next batch of model printing plans, the flexible layer can be thickened, the material hardness can be changed, or support structures can be added to the thin-walled areas or areas prone to stress.

[0093] Specifically, the operation feelings of medical personnel are collected, including feedback on the adaptability of different intubation equipment, the rationality of mechanical resistance, and the degree of restoration of structural details. If certain patient types or pathological features cannot be adequately simulated, further refinement of the accuracy or material allocation of the corresponding parts can be considered in the construction of the airway segmentation model and the production process of the head and neck model. Through continuous iteration and improvement, the personalized head and neck model can play a stronger clinical value in teaching and preoperative planning.

[0094] In other embodiments, during intubation training, pressure or strain sensors can also be embedded inside the head and neck model to record intubation mechanical data in real time for medical personnel to review; combined with AR / VR technology, the synchronized segmentation model and airway anatomy can be displayed in wearable devices for medical personnel to interact and practice with the physical model; using replaceable lesion modules such as tumors and stenosis bands, different pathologies can be quickly switched and simulated on the same head and neck model.

[0095] In summary, the head and neck model preparation method with difficult airway and the difficult airway intubation training method provided by the embodiment have the following technical effects: (1) more accurate anatomical details: fine identification of bony tissue and soft tissue is achieved by means of deep learning, and the restoration degree of the head and neck model in terms of nasal curvature, throat details, tracheal curvature and the like is significantly improved, making up for the neglect of fine structures by traditional two-dimensional or simplified three-dimensional models. (2) Better simulation and operation feeling: through multi-material partition printing, the difference in elasticity and hardness of soft tissue and bony tissue is fully reflected, and medical personnel can obtain a mechanical touch similar to that of a real person during intubation training, thereby obtaining a more realistic operation experience. (3) Standardized production and personalized customization: from the early image acquisition to the late printing forming, a modular process design is adopted, which can not only quickly generate general models in batches through standardized methods, but also can realize special customization according to individual pathologies. (4) Wider clinical application: in anesthesia, first aid, otolaryngology and teaching training and other scenes, the embodiment can provide different degrees of difficult airway simulation support, which not only improves the intubation success rate and safety of medical personnel, but also provides a more targeted solution for preoperative assessment and operation practice of special pathological cases.

[0096] It should be noted that in this document, the terms "comprising", "containing", or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device that includes a series of elements not only includes those elements, but also includes other elements not explicitly listed, or inherent to such a process, method, article or device. Without more limitations, the element defined by the statement "comprising a" does not exclude the presence of additional identical elements in the process, method, article or device that includes the element. In the description of the invention, the meaning of "a plurality of" is two or more, unless otherwise specified and limited.

[0097] In the description of the invention, unless otherwise specified and limited, the terms "mounting", "connection", "connection", "fixing" should be understood in a broad sense, for example, it can be fixed connection, or detachable connection, or integral; it can be mechanical connection, or electrical connection; it can be directly connected, or indirectly connected through an intermediate medium; it can be the internal communication of two elements or the interaction relationship between two elements. For those skilled in the art, the specific meaning of the above terms in the invention can be understood according to the specific circumstances.

[0098] While the application has been described in detail and with reference to specific preferred embodiments thereof, it will be apparent to one skilled in the art that various modifications and alternatives can be employed without departing from the spirit and scope of the application. Accordingly, the scope of the application should be determined by the appended claims and their equivalents.

Claims

1. A method for preparing a head and neck model with difficult airway, characterized in that: include: Obtain the first three-dimensional CT image with a difficult airway; preprocessing the first three-dimensional CT image to obtain a preprocessed first three-dimensional CT image; Annotating key anatomical regions on the preprocessed first three-dimensional CT image, and generating first three-dimensional mask data according to the annotation results; An airway segmentation model is constructed based on a deep neural network, and the airway segmentation model is trained using the preprocessed first three-dimensional CT image and the corresponding first three-dimensional mask data to obtain a trained airway segmentation model; Performing a three-dimensional CT scan on a patient with a difficult airway who is awaiting surgery to obtain a second three-dimensional CT image; preprocessing the second three-dimensional CT image and inputting the image into a trained airway segmentation model to obtain second three-dimensional mask data; A head and neck model of a patient with a difficult airway to be operated on is prepared based on the second three-dimensional mask data and using a 3D printing process.

2. The method for preparing a head and neck model with difficult airway according to claim 1, wherein: The preprocessing includes: noise reduction processing, grayscale normalization processing and cropping processing; and the preprocessed first three-dimensional CT image and the preprocessed second three-dimensional CT image only retain the head and neck area.

3. The method for preparing a head and neck model with difficult airway according to claim 1, wherein: The key anatomical regions include soft tissue and bony tissue; and the soft tissue includes the nasal cavity, oral cavity, pharynx, laryngeal cavity, and trachea; The nasal cavity includes the nasal conchae and the nasal septum; The oral cavity includes the tongue, hard palate and soft palate; The pharynx includes the pharyngeal wall and epiglottis; The laryngeal cavity includes the larynx, glottis and vocal cords; The bony tissue includes the skull base, mandible and cervical vertebrae.

4. The method for preparing a head and neck model with difficult airway according to claim 1, wherein: The deep neural network is a 3D U-Net convolutional neural network or a generative adversarial network.

5. The method for preparing a head and neck model with difficult airway according to claim 1, wherein: The steps of preparing a head and neck model of a patient with a difficult airway undergoing surgery based on the second three-dimensional mask data and using a 3D printing process include: Converting the second three-dimensional mask data into an STL format file; Use 3D modeling software to divide the bone tissue in the STL format file into rigid areas and the soft tissue into flexible areas; Printing the rigid region and the flexible region using a 3D printer; The printed rigid region and flexible region are assembled to obtain the head and neck model.

6. The method for preparing a head and neck model with difficult airway according to claim 5, wherein: The nasal concha, glottis and tongue in the flexible area are printed in blocks.

7. The method for preparing a head and neck model with difficult airway according to claim 5, wherein: The printing material of the rigid area is photosensitive resin or nylon.

8. The method for preparing a head and neck model with difficult airway according to claim 5, wherein: The printing material of the flexible area is elastic material.

9. The method for preparing a head and neck model with difficult airway according to claim 8, wherein: The printing material of the nasal cavity and trachea in the flexible area is transparent silicone.

10. A difficult airway intubation training method, characterized in that: include: The head and neck model is prepared by the method according to any one of claims 1 to 9; Mounting the head and neck model on an adjustable support frame; Disposing detachable pads at the nasal entrance and oral entrance of the head and neck model; Intubation training was performed on the head and neck model using an intubation device.

Images