Visual and adjustable artificial intelligence endotracheal intubation equipment

Abstract

The invention discloses a visually adjustable artificial intelligence endotracheal intubation device, comprising an endotracheal tube and an insertion tube inserted into the endotracheal tube, the insertion tube comprising an adjustment section, a connection section and a fixed straight section, the outer side of the fixed straight section is sleeved with a hard sleeve; the adjustment section is provided with a plurality of movable joints, the plurality of movable joints change direction through traction lines one and two, one end of the traction lines one and two are both fixed in the movable joints, and the other ends respectively pass through the side walls of the connection section and the fixed straight section in sequence and are then limited to the outer side of the insertion tube; after insertion, the position of the glottis of a patient is observed through a camera, the angle of the adjustment section is adjusted and the position of the glottis of the patient is inserted, and the glottis is automatically detected through a neural network model, so that the angle of entry of the front end can be operated more conveniently, the success rate of endotracheal intubation is improved, the incidence of adverse events such as hypoxia and death caused by endotracheal intubation failure is reduced, and at the same time, it is convenient for doctors to operate and reduce damage to patients.

Description

Technical Field

[0001] The invention relates to the field of intubation catheters, and in particular to a visually adjustable artificial intelligence tracheal intubation device. Background Art

[0002] The endotracheal tube is one of the medical devices that must be used in emergency treatment of respiratory arrest and general anesthesia in clinical surgery. When performing endotracheal intubation, a variety of intubation tools are needed to enable the endotracheal tube to accurately enter the patient's glottis and reach the trachea.

[0003] In the prior art, copper wire, aluminum wire and hard plastic are often used as the shaping support of the soft catheter. Although these catheter cores are modulated into the required shape outside the patient's mouth, they are prone to deformation when entering the patient's mouth. When the catheter is inserted into the glottis position in the patient's trachea after deformation, it is difficult to operate the catheter, and it takes multiple attempts to enter the glottis, increasing the risk of hypoxia and perioperative death of the patient. At the same time, it is difficult for doctors to operate, and it will also cause damage to the patient. It is especially not suitable for patients with abnormal airway, such as small mandible, difficulty opening the mouth, high glottis, infant epiglottis, throat tumors, laryngeal papilloma, tracheal deformity, etc.

[0004] In addition, it is also a huge challenge to quickly and accurately find the glottis position, especially for patients with the above-mentioned special diseases or special physiological structures, which often requires experienced doctors to complete. In recent years, with the development of robot automation and artificial intelligence technology, more intelligent and automated intubation equipment has also received attention and research and development in the field of tracheal intubation. The Kepler Intubation System (KIS) developed by Canada in 2012 is the first robotic system for tracheal intubation. The automated endoscopic tracheal intubation device (REALITI) based on laryngeal image recognition technology developed by the University of Zurich in Switzerland uses image recognition to detect the glottis opening. At the same time, some people have proposed using a neural network model to identify the glottis position. However, most of them are for fixed tracheal intubation equipment. For adjustable tracheal intubation equipment, there is not only longitudinal movement along the trachea, but also rotation operation approximately perpendicular to the trachea during the adjustment process of the intubation equipment. Therefore, the image contrast, target distance, target size, etc. change rapidly, making the image unclear, which is not conducive to quickly locking the glottis target. In addition, during the adjustment process, sputum and airway secretions will also partially block the lens, thereby interfering with positioning. Moreover, due to the complexity of the model, the traditional neural network algorithm has a poor response speed, which is extremely unfavorable for rapid adjustment and intubation operations, and will bring great safety risks and affect the intubation efficiency. This makes it difficult to actually apply the neural network model to clinical intubation operations. Summary of the invention

[0005] The purpose of the present invention is to solve the problems of the background technology and to propose a visually adjustable artificial intelligence endotracheal intubation device and a fast and accurate glottis positioning method used in conjunction with the adjustable endotracheal intubation device.

[0006] In order to achieve the above-mentioned purpose, the present invention adopts the following technical scheme: a tracheal tube 3 and an insertion tube 1 inserted into the tracheal tube 3,

[0007] Insertion tube 1, the front end of the insertion tube 1 close to the patient is provided with a camera and a cold light source, and is connected to an external device through a connecting connector at the other end.

[0008] The insertion tube 1 is integrally formed, and is divided into an adjustable angle section 11, a connecting section 12 and a fixed straight section 13 from the front section to the back. A hard metal sleeve 131 is sleeved on the outer side of the fixed straight section 13.

[0009] The adjustment section 11 is provided with a plurality of movable joints, and the plurality of movable joints change direction through the traction line 111 and the traction line 2 112. One end of the traction line 111 and the traction line 2 112 are fixed in the movable joints, and the other ends respectively pass through the side walls of the connecting section 12 and the fixed straight section 13 in sequence and are limited to the outside of the insertion tube 1. The method for determining the glottis during the intubation and adjustment process includes:

[0010] Step 1: Collect intubation image data for preprocessing;

[0011] Step 2: Build a deep neural network model and send the preprocessed image into the model for processing to determine whether the glottis appears;

[0012] The neural network model is as follows: the 1st to 5th layers are classic deep neural network structures, which perform convolution-pooling-convolution-pooling-convolution processing respectively; the 6th to 13th layers are independent heterogeneous fusion layers, and the output layer is used to output the result of glottis determination; the specific structure of the 6th to 13th layers is as follows:

[0013] ;

[0014] in is the output of the i-th hidden layer, is the output of the i-1th hidden layer, is the activation function, is the linear coefficient, is the linear intercept;

[0015] ;

[0016] set up , , are all 3*3 matrices, and All elements of are constants , is the coordinate relative to the center of the convolution kernel, Represents the phase angle.

[0017] ;

[0018] When i=6, ; when i=7, ; when i=8, ; when i=9, ; when i=10, ; when i=11, ; when i=12, ; when i=13, .

[0019] Furthermore,

[0020] One end of the pulling wire 111 and the pulling wire 2 112 located outside the insertion tube 1 is fixed on the winding column 41 of the screw cap 4 .

[0021] Furthermore,

[0022] The screw cap 4 is provided with an anti-rebound structure, including an elastic protrusion on the inner wall of the screw cap 4 and a non-return step on the outer side of the insertion tube 1 .

[0023] Furthermore,

[0024] The traction line 1 111 and the traction line 2 112 are arranged on the same plane; one end of the traction line 1 111 and the traction line 2 112 are fixed in the movable joint at the top, and the other end is connected and located outside the tracheal tube.

[0025] Furthermore,

[0026] Optical fibers and electrical wires are connected to the inner wall of the insertion tube, and a type-c power cord for connecting to a power source is connected to the end of the insertion tube close to the operator.

[0027] Furthermore,

[0028] The cold light source is an LED cold light source.

[0029] Furthermore,

[0030] The length of the adjustment section is set to 2-3cm, and the length of the connection section is set to 2-3cm. The front end of the tube core can be adjusted upward or downward to accurately enter the glottis and improve the success rate of tracheal intubation.

[0031] Furthermore,

[0032] After the diagnosis is made that it is the glottis, the operating physician is informed through image display and / or sound reminder.

[0033] Furthermore,

[0034] The notification method also includes marking the glottis position and / or area range using a marking box on the display screen.

[0035] Furthermore,

[0036] The robot is guided to perform autonomous intubation based on the positioning of the glottis.

[0037] In summary, due to the adoption of the above technical solution, the invention points and beneficial effects of the present invention are:

[0038] 1. In the present invention, the bending of the connecting section of the insertion tube will provide shaping support for the endotracheal tube, so that it can adapt to the shape of the patient's trachea. After insertion, the position of the patient's glottis is observed by a camera, and the specific structure of the insertion tube adjustment is optimized. By adjusting the angle and position of the adjustment section in real time, the front end can be more conveniently operated for entry, which is convenient for doctors to operate and helps reduce damage to patients.

[0039] 2. The present invention uses a deep learning method to establish a visual data detection model, which is specifically optimized for adjustable tracheal intubation operations (longitudinal advance and retreat changes of the image) and adjustment operations (lateral rotation changes of the image), and adds an independent glottis feature enhancement detection layer (the sixth to thirteenth layers). Experimental results show that compared with the classical method, the method of the present invention reduces the number of network layers, improves the detection accuracy of the glottis during tracheal intubation and adjustment, and increases the response speed.

[0040] 3. Furthermore, the image preprocessing was optimized before deep neural network processing, and the optimization algorithm cooperated with the neural network to improve the detection accuracy and speed. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 A schematic diagram of the structure of a catheter core according to an embodiment of the present invention is shown;

[0042] Figure 2 A schematic cross-sectional view of a catheter core structure provided in an embodiment of the present invention is shown;

[0043] Figure 3 A schematic diagram of a cap screwing structure provided according to an embodiment of the present invention is shown;

[0044] Legend:

[0045] 1. Insertion tube; 11. Adjustment section; 111. Pull line 1; 112. Pull line 2; 12. Connection section; 13. Fixed straight section; 131. Hard cannula; 3. Tracheal tube; 4. Tighten the cap; 41. Winding column. DETAILED DESCRIPTION

[0046] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0047] 1. Visual and adjustable artificial intelligence tracheal intubation equipment structure

[0048] See also Figure 1-2 The present invention provides a technical solution: a visually adjustable artificial intelligence endotracheal intubation device, comprising an endotracheal tube 3 and an insertion tube 1 inserted into the endotracheal tube 3.

[0049] An insertion tube 1 is provided with a camera and a cold light source at the front end close to the patient, and is connected to an external device through a connecting joint at the other end.

[0050] The insertion tube 1 is integrally formed and is divided into an adjustable section 11 with an adjustable angle, a connecting section 12 and a fixed straight section 13 from the front section to the back. A hard sleeve 131 is sleeved on the outer side of the fixed straight section 13 .

[0051] The adjustment section 11 is provided with a plurality of movable joints, and the directions of the plurality of movable joints are changed by traction line 111 and traction line 2 112. One ends of the traction line 111 and the traction line 2 112 are fixed in the movable joints, and the other ends respectively pass through the side walls of the connecting section 12 and the fixed straight section 13 in sequence and are limited to the outside of the insertion tube 1.

[0052] The inner wall of the insertion tube 1 is connected with optical fibers and electric wires, and the end of the insertion tube 1 close to the operator is connected with a type-c power cord for connecting to a power source.

[0053] See also Figure 3 In order to facilitate the use and fixation of the traction line 111 and the traction line 2 112 , one end of the traction line 111 and the traction line 2 112 limited to the outside of the insertion tube 1 is fixed on the winding column 41 of the screw cover 4 .

[0054] The screw cap 4 is provided with an anti-rebound structure, including an elastic protrusion on the inner wall of the screw cap 4 and a non-return step on the outer side of the insertion tube 1 .

[0055] When in use, the insertion tube 1 is inserted into the inner wall of the endotracheal tube 3, and the angle of the connecting section 12 is adjusted to suit the patient's trachea. The insertion tube 1 will also shape the endotracheal tube. Then, after inserting the patient's trachea, the type-c power cord close to the operator is connected to the mobile phone or other electronic device to power the cold light source and the camera, and the patient's image is transmitted to the mobile phone or other electronic device. After finding the glottis, the traction line 111 and the traction line 112 are pulled to adjust. The traction line 111 and the traction line 112 move in the opening of the movable joint, and the distance between the two adjacent movable joints on one side of the traction line 111 is shortened, so that the adjustment section 11 can be adjusted in the up and down direction. The adjustment angle is plus or minus 60 degrees. Pulling the traction line 112 is the same as pulling the traction line 111.

[0056] The adjustment section 11 can be adjusted according to specific circumstances to better adapt to the patient's trachea angle and reduce damage to the patient.

[0057] In the process of adjusting the pulling line 1 111 and the pulling line 2 112, the specific operation is to rotate the screw cap 4, wind the excess pulling line 1 111 or the pulling line 2 112 around the winding column 41 of the screw cap 4, and then fix it with the anti-rebound structure.

[0058] In order to facilitate the operator's observation, the type-c power cable close to the operator can be connected to a mobile phone or other device to power the cold light source and camera, and transmit the patient's image to the mobile phone or other electronic devices.

[0059] In order to more conveniently operate the up and down movement of the adjustment section 11, the adjustment section 11 includes a plurality of movable joints, a traction line 111 and a traction line 2 112 connected in sequence; the traction line 111 and the traction line 2 112 are arranged on the same plane; the traction line 111 and the traction line 2 112 are fixedly connected to the movable joint close to the patient's end by biological glue, and the traction line 111 and the traction line 2 112 are connected together close to the operator's end.

[0060] It is worth noting that the movable joint is provided with openings for the traction line 1 111 and the traction line 2 112 to pass through.

[0061] The traction belt pulls the traction line 1 111 or the traction line 2 112, and the traction line 1 111 and the traction line 2 112 are connected to the active joint closest to the patient, and the traction line 1 111 and the traction line 2 112 move in the opening of the active joint, so that the adjustment section 11 can be adjusted in the up and down direction, and can be adjusted in multiple directions in combination with the rotating tracheal tube. The traction belt can facilitate the operator to operate the traction line 1 111 and the traction line 2 112 to adjust the direction of the adjustment section 11, so as to reach the patient's tracheal position more efficiently, and the traction belt can also better adapt to the patient's tracheal shape, reducing the patient's tracheal damage.

[0062] In order to achieve tracheal intubation under clear vision and avoid excessive temperature, the light source installed around the camera is an LED cold light source.

[0063] The length of the adjustment section 11 is set to 2-3 cm, and the length of the connection section 12 is set to 2-3 cm. Considering that the length of the adult trachea is 11-13 cm, the length of the common tracheal tube entering the trachea after being inserted through the glottis is 4-6 cm. Therefore, the adjustment section 11 and the connection section 12 are the corresponding lengths entering the trachea. Under the traction of the traction line 111 and the traction line 2 112, the front end of the tube core can be adjusted upward or downward to accurately enter the glottis, thereby improving the success rate of tracheal intubation.

[0064] 2. Automatic glottis localization method

[0065] Step 1: Collect intubation image data for preprocessing

[0066] During the intubation process, since the intubation equipment can be bent and adjusted, the viewing angle of the endoscopic image is relatively small, and the image contrast, target distance, target size, etc. change rapidly, making the endoscopic image unclear. Therefore, the directly collected image data is preprocessed to reduce image noise so that the algorithm can detect the glottis target as soon as possible.

[0067] The preprocessing method is as follows.

[0068] 1. Set Indicates the number of currently collected Frame image, Represents the image captured by the next frame. Using the existing Lucas-Kanade algorithm to calculate arrive The optical flow is recorded as , is a matrix of the same size as the image. The value at is:

[0069] ;

[0070] Among them, i x Represents the component of the optical flow in the x direction of the image, iy Represents the component of the optical flow in the y direction of the image; is a vector representing the optical flow in the image The amount on. are the position coordinates in the image.

[0071] 2. Calculate motion estimation distance parameters:

[0072] ;

[0073] 3. Calculate the motion estimation phase parameters:

[0074] ;

[0075] in, The value is 0. , . here The value can be further refined to 16th order, 32nd order or any order, but according to experimental tests, higher order value division does not help or even reduces the preprocessing effect, so it is set to the above 8th order according to the present invention. argmax represents the maximum value of the formula Get the value.

[0076] 4. Estimating parameters based on motion , , calculate the motion estimation matrix .

[0077] ;

[0078] Respectively represent the center of the matrix as the zero point, The respective position offset in the direction.

[0079] 5. Set represents the Fourier transform symbol, Represents the inverse Fourier transform operator, which calculates the preprocessed image based on the motion estimation matrix :

[0080] ;

[0081] Preprocessing completed.

[0082] In the actual detection stage, the preprocessed data can be directly fed into the neural network model; in the sample production stage, the samples need to be labeled after preprocessing, as follows:

[0083] After the preprocessing is completed, the image samples are manually annotated in the preprocessed image, the position and contour of the glottis are marked in the image, and the contour is made to form a closed curve.

[0084] Among them, the method of forming a closed contour curve is as follows.

[0085] During manual labeling, a specific color marker value, such as blue, is used to manually mark the location and contour of the glottis in the image. After labeling, the labeled image is converted to the HSI space, i.e., the hue, saturation, and brightness space, to better extract the outline of the marker. The conversion formula is:

[0086] ;

[0087] min means taking the minimum value among the three.

[0088] After the manually marked image is saved, the color shift may occur due to image compression, quantization, etc., so the color value of the contour in the marked image may be slightly different from the color value of the original mark. Therefore, after calculating the H channel of the image according to the above formula, the contour mark value is recorded as , take the offset , when the value of the H channel falls into When , it is marked as 1 to indicate the contour, otherwise, it is marked as 0. This generates a contour extraction binary image .

[0089] To ensure The closure of the binary image is further subjected to morphological connection operations, namely:

[0090] ;

[0091] In the above formula Representing images China-Israel A local area centered on The above formula means that if the coordinates are A local area centered on In the example, all pixels have a value of 1, then Corresponding position The pixel value at which the pixel is located is 1. Otherwise, the pixel value is 0.

[0092] Further, remember:

[0093] ;

[0094] In the above formula, Representing images China-Israel A local area centered on The above formula means that if the coordinates are A local area centered on In the case where there is a pixel with a value of 1, then Corresponding position The pixel value at which the pixel is located is 1. Otherwise, the pixel value is 0.

[0095] By controlling the above local areas , The size of the processed contour can ensure that it is closed.

[0096] According to the manually annotated glottis position contour, the circumscribed rectangle of the closed contour is taken as the positive sample.

[0097] A rectangular area is randomly selected in the image, which has no intersection with the rectangular area of ​​the positive sample mentioned above, as a negative sample.

[0098] Through the preprocessing of the above samples and the cooperation with the following neural network model, it is ensured that the training results of the neural network model are more suitable for the adjustable endotracheal tube to quickly and accurately find the glottis during the intubation and adjustment process.

[0099] Step 2: Build a deep neural network model for adjustable endotracheal tube equipment and perform glottis detection

[0100] A visual data learning and detection model using a deep neural network model uses the positive samples and negative samples obtained in step 1 to learn model parameters to detect and obtain the position of the glottis in the image.

[0101] The image of the input model is recorded as , represents the image of the original image data collected by the endoscope after preprocessing in step 1. Indicates that The pixel value at position.

[0102] A deep neural network model is proposed, which is described as follows.

[0103] First hidden layer:

[0104] ;

[0105] The output of the first hidden layer are several feature maps, including There are 32 independent convolution kernels with a size of 5*5, which are used to model the local features of the glottis. , Represents the linear coefficient and linear intercept of the convolution kernel, is the activation function of the neural network, Refers to the coordinates relative to the center of the image, Refers to the coordinates relative to the center of the convolution kernel.

[0106] Activation functions for neural networks Defined as:

[0107] ;

[0108] In the formula, is the scale parameter of the activation function, which is used to adjust the learning convergence speed. Arctan represents the inverse tangent function. Compared with the classic sigmoid function, ReLU and other functions, the above piecewise function has a better classification effect on glottal samples. The value is taken as 0.005 based on experience.

[0109] Second hidden layer:

[0110] ;

[0111] The output of the second hidden layer It is a number of downsampled feature maps, corresponding to the outputs of the first hidden layer. Max represents maximum pooling, that is, taking the local maximum value in the 4*4 window.

[0112] Third hidden layer:

[0113] ;

[0114] The output of the third hidden layer is a lower-level feature map, including There are 32 independent convolution kernels in total, and the convolution kernel size is 3*3. , Represents the linear coefficient and linear intercept of the third layer convolution kernel, is the activation function of the neural network, Refers to the coordinates relative to the center of the image, Refers to the coordinates relative to the center of the convolution kernel.

[0115] Fourth hidden layer:

[0116] ;

[0117] Fourth hidden layer output It is the feature map after further downsampling, corresponding to the outputs of the third hidden layer, and max represents maximum pooling.

[0118] Fifth hidden layer:

[0119] ;

[0120] Fifth hidden layer output is the lowest level feature map, including There are 32 independent convolution kernels in total, and the convolution kernel size is 3*3. , Represents the linear coefficient and linear intercept of the fifth layer convolution kernel, is the activation function of the neural network, Refers to the coordinates relative to the center of the image, Refers to the coordinates relative to the center of the convolution kernel.

[0121] The classic deep convolutional neural network (i.e., the first five layers) realizes the mining and extraction of various local features that describe the appearance of the target by setting multiple hidden convolutional layers at different levels, thereby realizing the classification of specific targets. In general image classification applications, the ability to express the appearance of the target can be improved by setting more convolutional layers. For example, YOLOv3 contains more than 70 convolutional layers. However, a higher number of layers increases the number of parameters and reduces the computational efficiency of learning and detection. In this example, through the analysis of the appearance characteristics of the glottis target in the endoscopic image and experimental testing, a deep neural network based on the glottis feature enhancement detection layer is proposed, which reduces the number of network layers and improves the detection efficiency of the glottis target.

[0122] By performing heterogeneous fusion on the local convolution features (Equations 1-6) of each level obtained by the classic convolutional network, that is, setting convolution kernels of different phases to further model the feature quantities obtained by the convolutional network, the parameters of each heterogeneous layer are obtained as a supplement to the classic convolutional layer, which greatly reduces the number of convolutional layers and obtains excellent detection performance. The heterogeneous layers are as follows.

[0123] Sixth hidden layer:

[0124] ;

[0125] The sixth hidden layer output , input part is the activation function, is the linear coefficient, is the linear intercept, where:

[0126]

[0127] set up , , are all 3*3 matrices, and All elements of are constants , is the coordinate relative to the center of the convolution kernel, , The matrix elements of can be obtained by neural network training. Represents the phase angle.

[0128] ;

[0129] In the sixth hidden layer, the phase angle is taken according to formula 8 .

[0130] Seventh hidden layer:

[0131] ;

[0132] The seventh hidden layer output , input part is the activation function, is the linear coefficient, is the linear intercept, The value is the same as the right side of equation 8, but the phase angle is .

[0133] Eighth hidden layer:

[0134] ;

[0135] The output of the eighth hidden layer , input part is the activation function, is the linear coefficient, is the linear intercept, The value is the same as the right side of equation 8, but the phase angle is .

[0136] Ninth hidden layer:

[0137] ;

[0138] Ninth hidden layer output , input part is the activation function, is the linear coefficient, is the linear intercept, The value is the same as the right side of equation 8, but the phase angle is .

[0139] Tenth hidden layer:

[0140] ;

[0141] Tenth hidden layer output , input part is the activation function, is the linear coefficient, is the linear intercept, The value is the same as the right side of equation 8, but the phase angle is .

[0142] Eleventh hidden layer:

[0143] ;

[0144] The eleventh hidden layer output , input part is the activation function, is the linear coefficient, is the linear intercept, The value is the same as the right side of equation 8, but the phase angle is .

[0145] Twelfth hidden layer:

[0146] ;

[0147] The output of the twelfth hidden layer , input part is the activation function, is the linear coefficient, is the linear intercept, The value is the same as the right side of equation 8, but the phase angle is .

[0148] Thirteenth hidden layer:

[0149] ;

[0150] Thirteenth hidden layer output , input part is the activation function, is the linear coefficient, is the linear intercept, The value is the same as the right side of equation 8, but the phase angle is .

[0151] The sixth to thirteenth hidden layers capture the phase parameters of the glottal features to be identified through multiple independent heterogeneous fusion layers. Compared with the classic convolutional network (such as YOLOv3), the number of network layers is reduced, which is more suitable for adjustable endotracheal intubation equipment and improves the recognition performance during intubation and adjustment.

[0152] Fourteenth hidden layer:

[0153] ;

[0154] The output of the fourteenth hidden layer , is the linear coefficient, is the linear intercept, is the activation function.

[0155] Fifteenth hidden layer:

[0156] ;

[0157] The output of the fifteenth hidden layer is a fully connected layer, is the linear coefficient, is the linear intercept, is the activation function.

[0158] The final output layer is defined as:

[0159] ;

[0160] is the linear coefficient, is the linear intercept, is the activation function. Output Indicates whether the input data corresponds to the glottal target.

[0161] The positive and negative sample data obtained in step 1 are used to train the above model, and the cost function is defined as:

[0162] ;

[0163] represents the training sample label value, Represents the model output value.

[0164] After the above model training is completed, the parameters of each linear coefficient and linear intercept are obtained, and the input test sample is classified according to the trained parameters to obtain whether the input image is a glottal target. The specific experimental data is as follows:

[0165]

[0166] After the algorithm determines that it is the glottis, the operating doctor can be informed through image display, sound reminder, etc., thereby improving the safety and accuracy of the operation and reducing misoperation. For example, the glottis position and / or area range can be marked with a label box on the display screen.

[0167] It is understandable that the present invention may also be combined with an automatic intubation robot, and the robot may be guided to perform autonomous intubation operations based on the positioning of the glottis.

[0168] The above are only preferred specific implementation modes of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical solutions and inventive concepts of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A system combining a visually adjustable artificial intelligence tracheal intubation device and an automatic intubation robot, characterized in that: It comprises a tracheal tube (3) and an insertion tube (1) inserted into the tracheal tube (3); An insertion tube (1), wherein a camera and a cold light source are arranged at the front end of the insertion tube (1) close to the patient, and the insertion tube (1) is connected to an external device via a connecting joint at the other end; The insertion tube (1) is integrally formed and is divided into an adjustable angle section (11), a connecting section (12) and a fixed straight section (13) from the front section to the back, and a hard metal sleeve (131) is sleeved on the outer side of the fixed straight section (13); The adjustment section (11) is provided with a plurality of movable joints, and the plurality of movable joints change direction through a traction line 1 (111) and a traction line 2 (112); one end of the traction line 1 (111) and the traction line 2 (112) are fixed in the movable joints, and the other ends respectively pass through the side walls of the connection section (12) and the fixed straight section (13) in sequence and are then limited to the outside of the insertion tube (1); The robot is guided to perform autonomous intubation based on the positioning of the glottis; The glottis determination methods during intubation and adjustment include: Step 1: Collect intubation image data for preprocessing; Step 2: Build a deep neural network model and send the preprocessed image into the model for processing to determine whether the glottis appears; The neural network model is as follows: the 1st to 5th layers are classic deep neural network structures, which perform convolution-pooling-convolution-pooling-convolution processing respectively; the 6th to 13th layers are independent heterogeneous fusion layers, and the output layer is used to output the result of glottis determination; the specific structure of the 6th to 13th layers is as follows: ； in is the output of the i-th hidden layer, is the output of the i-1th hidden layer, is the activation function, is the linear coefficient, is the linear intercept; ； set up , , are all 3*3 matrices, and All elements of are constants , is the coordinate relative to the center of the convolution kernel, represents the phase angle; ； When i=6, ; when i=7, ; when i=8, ; when i=9, ; when i=10, ; when i=11, ; when i=12, ; when i=13, ; After the diagnosis is made that it is the glottis, the operating doctor is informed through image display and sound reminder; The notification method also includes marking the glottis position and / or area range using a marking box on the display screen.

2. The system combining a visually adjustable artificial intelligence tracheal intubation device and an automatic intubation robot according to claim 1, characterized in that: One end of the pulling wire 1 (111) and the pulling wire 2 (112) located outside the insertion tube (1) is fixed on a winding column (41) of the screw cap (4).

3. The system combining a visually adjustable artificial intelligence tracheal intubation device and an automatic intubation robot according to claim 2, characterized in that: The twist-off cover (4) is provided with an anti-rebound structure, comprising an elastic protrusion on the inner wall of the twist-off cover (4) and a non-return step on the outer side of the insertion tube (1).

4. The system combining a visually adjustable artificial intelligence tracheal intubation device and an automatic intubation robot according to claim 1, characterized in that: The traction line 1 (111) and the traction line 2 (112) are arranged on the same plane; one end of the traction line 1 (111) and the traction line 2 (112) are fixed in the movable joint at the top, and the other end is connected to and located outside the tracheal tube (3).

5. The system combining a visually adjustable artificial intelligence tracheal intubation device and an automatic intubation robot according to claim 1, characterized in that: The inner wall of the insertion tube (1) is connected with optical fibers and electric wires, and the end of the insertion tube (1) close to the operator is connected with a type-c power cord for connecting to a power source.

6. The system combining a visually adjustable artificial intelligence tracheal intubation device and an automatic intubation robot according to claim 1, characterized in that: The cold light source is an LED cold light source.

7. The system combining a visually adjustable artificial intelligence tracheal intubation device and an automatic intubation robot according to claim 1, characterized in that: The length of the adjustment section (11) is set to 2-3 cm, and the length of the connection section (12) is set to 2-3 cm.

Images