Intelligent training system for sputum suction operation

Abstract

The invention discloses an intelligent training system for sputum suction operation. The intelligent training system comprises a simulated human body, a sputum suction tube and a monitoring device. The outermost airway and the inner detachable simulation part attached to the airway are arranged in the trachea of the simulated human body, and different types of simulation parts are arranged, so that simulation of patients with various illness states and various physiological conditions is achieved, and training comprehensiveness is guaranteed. Through the monitoring device and a special algorithm arranged in the monitoring device, automatic sputum suction exercise can be carried out without guarding of teachers.

Description

Technical Field

[0001] The present invention relates to the field of medical devices, and in particular to an intelligent training system for sputum suction operation. Background Art

[0002] Suctioning is a routine operation for most bedridden patients and patients with respiratory tract problems. However, since patients usually have fragile respiratory tracts, the suctioning operation technique requires high skills. Therefore, a suctioning operation training device is urgently needed.

[0003] The existing technology is to fill simulated sputum in a simulated airway, but usually because the airway is tubular, the simulated sputum fills a certain area of ​​the airway, while the actual sputum may block the entire airway, or it may stick to the airway wall. Therefore, the simulation of the existing technology does not match the actual situation and cannot achieve real training.

[0004] The existing simulated airways are mainly straight lines, and at the bends of the airways, there are only two tubular airways connected at a certain geometric angle. This is different from the smooth curve of the actual human airway. In addition, the real airway will have certain bends even in places that are close to straight lines. In addition, the airway conditions of each person are different, and there may be some physiological protrusions or depressions in certain positions of the airway. How to simulate this situation realistically is also a challenge.

[0005] In addition, even if the real airway is simulated, a simulated human body can only simulate one airway structure, which is far from enough for practitioners. Therefore, how to enable practitioners to conveniently and quickly practice airways with different structures and different pathologies, and simulate suction operations under different patient individuals, different diseases, and different physiological conditions is also an urgent problem to be solved.

[0006] In the prior art, the mannequin is usually skin-colored, which makes it impossible for the operator to see the airway condition. Although it is suitable for assessment, it is not suitable for practice. Moreover, most of the current exercises are supervised by teachers, which requires teacher resources. When the teacher is not present, the practitioner cannot judge whether his operation is correct. There is currently no automatic evaluation method to measure the operator's operation. Although some people have proposed the use of pressure sensors, it is impossible to simulate the real situation with only a limited number of sensors. Real patients have receptors in every position, and the operator needs to be extra careful in every step. The existing simulation device obviously cannot achieve this function with a limited number of sensors. Summary of the invention

[0007] In view of the deficiencies in the prior art, the present invention provides an intelligent training system for sputum suction operation, which solves the problems raised in the above-mentioned background technology.

[0008] To achieve the above objectives, the present invention is implemented through the following technical solutions:

[0009] An intelligent training system for suction operation, simulating a human body, a suction tube, and a monitoring device;

[0010] The simulated human body includes the outer surface, lungs and trachea;

[0011] The surveillance device includes a camera and a field processor;

[0012] The trachea is composed of the outermost airway and the simulation part which fits the airway from the outside to the inside. The outer wall of the simulation part has a protruding buckle, and the inner wall of the airway has a groove. Through the cooperation of the buckle and the groove, the simulation part can be detachably installed inside the airway.

[0013] The outer wall of the simulation part is cylindrical and fits the inner wall of the airway, and the inner wall of the simulation part is an irregular curved surface, including at least one or more physiological curved parts, one or more mass parts protruding inwardly, and one or more narrow parts;

[0014] The trachea can be opened along the horizontal cross section of the simulated human body lying flat, and divided into upper and lower parts, thereby exposing the horizontal cross section of the airway and the simulated part;

[0015] The sputum is set in the space surrounded by the inner wall of the simulation part.

[0016] The outer surface of the simulated human body is made of transparent resin.

[0017] The outer surface of the simulated human body can be opened along a horizontal cross section when the simulated human body is lying flat, thereby exposing the lungs and the trachea.

[0018] There are various types of simulation departments, including conventional, pathological and special types.

[0019] The simulation unit's production method is to perform three-dimensional modeling through CT and MRI images of the trachea of ​​multiple patients, and then perform 3D printing.

[0020] Sputum is composed of a gel.

[0021] The sputum is arranged to adhere to the inner wall of the simulation part 5 .

[0022] The sputum is arranged to gather in the channel surrounded by the inner wall of the simulation part.

[0023] A large amount of sputum is set to be congested in the channel surrounded by the inner wall of the simulation part.

[0024] The camera of the monitoring device collects video, the on-site processor receives the video, and the intelligent training and discrimination method for suction operation is run in the on-site processor, so as to automatically discriminate the suction operation and remind the operator.

[0025] The present invention has the following inventive features and beneficial effects:

[0026] 1. The detachable settings of the trachea and simulation part, as well as the irregular design inside the simulation part and the design of different types of structures (such as ulcers, masses, cysts, damage, airway width, special bends), make it easy to replace the simulation part, so as to quickly switch the practice content, so that the practitioner can face a variety of different physiological conditions, pathological conditions, and individual patient conditions more realistically, and practice suction operations under the above-mentioned various conditions, making the practice more comprehensive and sufficient.

[0027] 2. The design that the trachea and the simulation part can be opened along the horizontal section makes it possible to freely place sputum in the simulation part, thereby simulating the clinical situation of sputum sticking to the airway and achieving more complete training.

[0028] 3. The monitoring device can automatically monitor the practitioner's suction action, and through the processing of a specially designed neural network algorithm, the practitioner's action can be judged and prompted whether it is standard or not, thereby helping the practitioner to practice independently and achieving the above functions more accurately and at a low cost.

[0029] Instruction Manual

[0030] Figure 1 This is a schematic diagram of the overall structure of the simulated human body of the present invention;

[0031] Figure 2 The schematic diagram of the internal structure of the horizontal cross section of the trachea simulating the human body of the present invention; DETAILED DESCRIPTION

[0032] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0033] A sputum suction operation intelligent training system comprises a simulated human body, a special sputum suction tube and a monitoring device.

[0034] The simulated human body includes an outer surface composed of transparent resin simulating human skin, etc., to ensure that the operator can see the internal situation and facilitate monitoring equipment to detect. Figure 1 As shown, the outer surface of the simulated human body can be opened along a horizontal cross section when the simulated human body is lying flat, so that the lungs 1 and the trachea 2 are exposed.

[0035] like Figure 2 As shown, the trachea 2 can also be opened along the horizontal cross section of the simulated human body lying flat. From the outside to the inside, the trachea 2 is the outermost airway 6, the replaceable simulation part 5 connected to the airway 6, and the sputum 9 attached to the simulation part 5.

[0036] Among them, the outer wall of the simulation part 5 has a protruding buckle 4, and the inner wall of the airway 6 has a groove 8. Through the cooperation of the buckle 4 and the groove 8, the simulation part 5 can be detachably installed inside the airway 6. In this way, not only can the simulation part be replaced conveniently, but also the simulation part and the airway are ensured to be firmly connected to each other, avoiding sliding during operation, and more in line with actual operation. This is also one of the inventive points of the present invention. At present, the existing devices inside the airway are all set in the airway. During the insertion of the suction tube, the relative positions of the devices inside the airway will be destroyed due to the operator's unskilled operation, thereby reducing the service life. And a completely fixed device makes replacement difficult.

[0037] The outer wall of the simulation part 5 is cylindrical and fits the inner wall of the airway 6. The inner wall of the simulation part 5 is an irregular curved surface, which at least includes one or more physiological curved parts 3, one or more inwardly protruding mass parts 7, and one or more narrow parts 10. Since the simulation part 5 is connected to the airway 6, different simulation parts 5 can be replaced to achieve different training purposes. For example, a simulation part with a larger physiological curve can be installed to practice how to perform sputum suction in this special patient situation; sputum 9 can be placed at the distal end of the mass part 7 to create obstacles to sputum suction practice, so as to practice sputum suction operation in this situation, and so on.

[0038] Understandably, Figure 2 The examples are only used to illustrate the special structures that may exist on the inner wall of the simulation part 5: the mass part 7, the stenosis part 10, and the physiological curvature part 3. In fact, different inner walls of the simulation parts 5 may also have ulcers, masses, cysts, damage, etc., or their structures may have airway width, special curvature, etc. These can all be designed according to actual needs. In particular, the simulation part 5 can also be used to simulate deformities in special cases. Therefore, the simulation part can have various types, including conventional type (normal physiological structure), pathological type (with ulcers, masses, cysts, damage, etc.) and special type (with special physiological structure, such as airway width, special curvature, etc.). The production method of the simulation part 5 can be to perform three-dimensional modeling through CT and MRI images of the trachea of ​​multiple patients, and then perform 3D printing to achieve it.

[0039] The trachea 2 can also be opened along the horizontal cross section of the simulated human body lying flat, and divided into upper and lower parts, thereby exposing the horizontal cross section of the airway 6 and the simulation part 5, so as to facilitate the injection of sputum into different positions of the simulation part. The upper and lower parts are also matched by buckles and grooves to prevent the trachea 2 from sliding undesirably after the upper and lower parts are buckled together.

[0040] The sputum 9 is made of gel. The sputum can adhere to the inner wall of the simulation part 5 to simulate the sticky situation; it can also gather in the channel surrounded by the inner wall of the simulation part 5 to simulate the sputum congestion; in particular, since the sputum is added manually after the airway 6 is opened, the amount of sputum can be added arbitrarily to simulate the situation of a large amount of sputum congestion. In order to simulate the sputum that is difficult to remove, the inner wall surface of the simulation part 5 can be provided with tiny protrusions to increase the adsorption force of the sputum gel.

[0041] The outer surface of the simulated human body, the trachea 2, the airway 6, and the upper part of the simulation part 5 are all made of transparent materials, which is convenient for the practitioner to visually observe the operation of the suction tube, thereby practicing the suction operation technique. In addition, it is also convenient for the monitoring device to monitor the suction operation process, thereby automatically judging whether the suction operation meets the requirements, especially when multiple special simulation parts are installed, so as to warn the operator and enable him to practice independently without other teachers present.

[0042] The inner wall of the lower half of the simulation part 5 can be set to a fixed color, for example, red. The sputum 9 can be set to green. The sputum suction tube is a commonly used sputum suction tube in clinical practice, but a mark can be set on it, such as a black ring mark. In this way, it is not only convenient for the operator to observe his own actions and correct his own operations during operation; it can also make it easy for the monitoring device to identify and distinguish during judgment, thereby improving the judgment accuracy and reducing the difficulty of the algorithm.

[0043] It is understandable that the above colors can be adjusted according to actual needs, for example, each part is adjusted to the color corresponding to the real human body.

[0044] The monitoring device comprises a camera arranged above the simulated human body; and also comprises a field processor for receiving and processing the video sent by the camera so as to judge the standardization of the sputum suction operation.

[0045] In order to save costs and facilitate on-site deployment, all algorithm processing is completed on-site in the on-site processor without using a remote server. This puts higher requirements on the algorithm, that is, to save computing power, and at the same time, the discrimination accuracy should meet the requirements. To this end, the present invention proposes an algorithm specifically for distinguishing sputum suction operations on a simulated human body, as follows:

[0046] The present invention innovatively proposes a suction behavior detection unit and method. A camera is used to capture the actions of the suction operator and the insertion position of the suction tube, and a field processor is used to identify the operator's operation position and strength online, and an alarm is issued when an operation error occurs.

[0047] Different from the current classic convolutional neural network structure, by introducing spatiotemporal optimization steps for suction behavior and operation object characteristics, the number of network layers is simplified, the number of network parameters is reduced, and thus the network's computational efficiency is improved.

[0048] At present, the behavior detection methods based on images / videos are usually based on two-dimensional convolutional networks superimposed with time series structures such as LSTM, or three-dimensional convolutional networks for local spatiotemporal modeling. The problem with the above methods is that the computational complexity is high, the number of model parameters is large, and it is not easy to run on edge devices with limited resources.

[0049] In order to reduce the complexity of the network, a spatiotemporal optimization method based on the characteristics of suction behavior and operation objects is proposed through a large number of research experiments. The suction behavior and the target image data of the operation object are feature mapped using the spatiotemporal optimization method, and the features with higher correlation with the suction behavior are extracted, thereby streamlining the number of layers and parameters of the classification network and achieving the goal of fast operation on edge devices.

[0050] One of the a priori characteristics of suction behavior is that the operation space is relatively limited, and the response level of local behavior is significantly higher than that of the surrounding area, which is manifested in that the samples are more concentrated in a narrow spatial scale range.

[0051] The second prior of suction behavior is that the action is slow in time, which is reflected in the stronger response of samples in areas with higher time scales.

[0052] Based on the above two observation priors, the present invention proposes a recognition method based on spatiotemporal optimization oriented to suction behavior and operation object characteristics.

[0053] Step 1: Video preprocessing. Expand the three-dimensional spatiotemporal space representing continuous image data to a five-dimensional scale space. By adding a time scale dimension and a space scale dimension, the nonlinear samples of the three-dimensional spatiotemporal space are mapped to linear samples of the five-dimensional scale space, so that the original linearly inseparable samples become linearly separable, thereby improving the recognition performance of the model.

[0054] Assume that an image data unit (i.e. a video) describing a behavior is represented as:

[0055] I(x,y,t)

[0056] x, y are spatial coordinates, and t is the time coordinate. The scale mapping is defined as follows:

[0057]

[0058] 0≤α<λ α ,0≤β<λ β And α ,λ β ,α,β are integers

[0059] In the scale space, according to the experimental results, the optimal scale range is 1≤λ α ≤8, 1≤λ β ≤32.

[0060] Step 2 is based on the prior one mentioned above. In the scale space, the optimal solution of the spatial scale is tested and selected. The process is as follows.

[0061] First, calculate the function

[0062]

[0063] The minimum value of .

[0064]

[0065] N is the total number of pixels in one image data unit.

[0066] Formula 2 means that, given λ α When 1≤λ β The value of σ that maximizes σ when ≤32.

[0067] Secondly, select the parameter λ that makes the function f1 take the minimum value α The value of As the optimal solution for spatial scale.

[0068] Step 3: Based on the optimal solution of spatial scale, a learning model in time scale space is established to classify the input data.

[0069] The input data is recorded as

[0070] in

[0071] In four-dimensional linear space As input, a neural network learning model is established.

[0072] The implementation of the first hidden layer of the model is as follows.

[0073]

[0074] Among them, ω 1i Each is an independent three-dimensional linear convolution kernel describing local features, τ 1i is the corresponding linear bias, the subscript i ranges from 1 to 256, indicating that the number of independent convolution kernels is 256, and ξ represents the activation function:

[0075]

[0076] θ is the control parameter, preferably The activation function of Formula 5 can improve the classification performance of time-scale data compared to classic activation functions such as Sigmoid and ReLU.

[0077] The implementation of the second hidden layer of the model is as follows.

[0078]

[0079] Where 0≤u,v<4,0≤w<8 are all integers. It can be seen that the second hidden layer is the maximum pooling layer. The spatial and temporal dimensions are pooled to reduce the original

[0080] The implementation of the third hidden layer of the model is as follows.

[0081]

[0082] ω 3i Each is an independent three-dimensional linear convolution kernel describing local features, τ 3i is the corresponding linear bias, the subscript i ranges from 1 to 256, indicating that the number of independent convolution kernels is 256, and τ represents the aforementioned activation function.

[0083] The implementation of the fourth hidden layer of the model is as follows.

[0084]

[0085] Where 0≤u,v<4,0≤w<8 are all integers. It can be seen that the fourth hidden layer is the maximum pooling layer. The spatial and temporal dimensions are pooled to reduce the original

[0086] After two convolutional poolings, the convolution coefficients preliminarily establish the local features of the image data units after mapping them to the optimal spatial scale.

[0087] Since the optimal spatial scale has been found in the previous steps, through experimental tests, after two layers of convolution, the extracted local features are sufficient to achieve a good behavior recognition effect. Therefore, two convolution layers are used. Compared with the existing convolutional network (hundreds of layers), the number of parameters is greatly reduced.

[0088] Furthermore, the implementation of the fifth hidden layer of the model is as follows.

[0089]

[0090] The fifth hidden layer maps the time scale differently according to the high and low scales based on the prior 2, where the low scale 1≤λ β ≤16 is mapped to the same node in the next layer, and the high scale is 17≤λ β ≤32 are mapped to the same number of nodes in the next layer to capture the characteristics of the high-scale part of time in more detail. is the linear mapping parameter of the fifth hidden layer, τ5 is the corresponding linear bias, and ξ represents the aforementioned activation function. By introducing this layer, it is possible to effectively capture the stronger response characteristics on the time scale corresponding to the behavior and improve the recognition and classification of the behavior.

[0091] The implementation of the sixth hidden layer of the model is as follows.

[0092]

[0093] The sixth hidden layer is a fully connected layer, where is a linear mapping parameter, τ6 is the corresponding linear bias, and n represents the output dimension of the sixth layer, preferably n is 256 dimensions.

[0094] The model output layer is implemented as follows.

[0095]

[0096] o represents the output behavior classification category. e is the natural exponential function. is the linear mapping coefficient, τ7 is the corresponding linear bias, and ξ is the activation function.

[0097] The output behavior classification categories include γ>0, indicating normal behavior, and γ<0, indicating abnormal behavior.

[0098] Compared with the existing methods, the optimized model of the method of the present invention has fewer layers. Compared with the existing neural network models with dozens or even hundreds of layers, the number of parameters is greatly reduced, the calculation efficiency is higher, and it has a considerable recognition accuracy. Therefore, the above neural network model is also one of the inventive points of the present invention and is an important component that can realize automatic sputum suction training.

[0099] The cost function used to train the above model is

[0100]

[0101] Among them, γ is the model output value, is the classification of training samples. According to Formula 11, the dimension output corresponding to the normal category is 1, and the abnormal dimension output is -1.

[0102] After training, the above model is used to recognize input data, and the largest output dimension is taken as the category of corresponding behavior recognition.

[0103] This paper proposes an innovative, more lightweight intelligent detection system for sputum suction behavior that can be deployed on edge computing devices to implement computational reasoning. The experimental results show that the behavior recognition rate is comparable to that of existing comparable methods, and it greatly improves the computational efficiency and reduces the computing power requirements. Under the same processor, the specific experimental results are as follows:

[0104]

[0105]

[0106] The above description is only a preferred specific implementation manner 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 scheme and inventive concept 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. An intelligent training system for sputum suction operation, characterized in that: Simulated human body, suction tube, monitoring device; The simulated human body includes the outer surface, lungs and trachea; The surveillance device includes a camera and a field processor; The trachea is composed of the outermost airway and the simulation part which fits the airway from the outside to the inside. The outer wall of the simulation part has a protruding buckle, and the inner wall of the airway has a groove. Through the cooperation of the buckle and the groove, the simulation part can be detachably installed inside the airway. The outer wall of the simulation part is cylindrical and fits the inner wall of the airway, and the inner wall of the simulation part is an irregular curved surface, including at least one or more physiological curved parts, one or more mass parts protruding inwardly, and one or more narrow parts; The trachea can be opened along the horizontal cross section of the simulated human body lying flat, and divided into upper and lower parts, thereby exposing the horizontal cross section of the airway and the simulated part; The sputum is set in the space surrounded by the inner wall of the simulation part.

2. The system according to claim 1, characterized in that: The outer surface of the simulated human body is made of transparent resin.

3. The system according to claim 1, characterized in that: The outer surface of the simulated human body can be opened along a horizontal cross section when the simulated human body is lying flat, so as to expose the lungs and trachea.

4. The system according to claim 1, characterized in that: There are various types of simulation departments, including conventional, pathological and special types.

5. The system according to claim 4, characterized in that: The simulation unit's production method is to perform three-dimensional modeling through CT and MRI images of multiple patients' trachea, and then perform 3D printing.

6. The system according to claim 1, characterized in that: Sputum is composed of a gel.

7. The system according to claim 6, characterized in that: The sputum is arranged to adhere to the inner wall of the simulation part 5 .

8. The system according to claim 6, characterized in that: The sputum is arranged to gather in the channel surrounded by the inner wall of the simulation part.

9. The system according to claim 6, characterized in that: A large amount of sputum is set to be congested in the channel surrounded by the inner wall of the simulation part.

10. The system according to claim 1, characterized in that: The camera of the monitoring device collects video, the on-site processor receives the video, and the intelligent training and discrimination method for suction operation is run in the on-site processor, so as to automatically discriminate the suction operation and remind the operator.

Images