Mobile infusion teaching experiment system based on digital twinborn technology

Abstract

The invention discloses a mobile infusion teaching experiment system based on a digital twinning technology, which collects data through an infusion operation module, and transmits the data to a digital twinning modeling module through a data collection and transmission module to construct a virtual model and perform real-time mapping. The mobile scene simulation module provides complex scene simulation, the real-time feedback module analyzes errors by using an artificial intelligence algorithm and performs multi-modal feedback, the intelligent evaluation module generates a skill evaluation report, the teacher management module supports teacher monitoring and management, and the database module stores data and assists in teaching optimization. The system improves the operation skill and risk coping ability of students, optimizes teaching management, and has a wide application prospect.

Description

Technical Field

[0001] The present invention relates to the cross - field of medical education and information technology, and specifically to a mobile infusion teaching experiment system based on digital twin technology. Background Art

[0002] Infusion occupies a fundamental and crucial position in medical nursing work. However, traditional infusion teaching faces many difficulties. On the one hand, the scarcity of teaching resources severely restricts students' practical opportunities. The limited model humans and infusion devices cannot meet the practice needs of many students, resulting in students having difficulty fully familiarizing themselves with the operation process and slow improvement in operation proficiency. On the other hand, traditional teaching lacks an effective real - time feedback mechanism. When students practice operations, errors cannot be pointed out and corrected in a timely manner, which is not conducive to accurately mastering operation skills. At the same time, teachers' assessment of students' skills mostly relies on subjective judgment, lacking objective and quantitative data support, and it is difficult to comprehensively and accurately measure students' real levels. In addition, traditional teaching methods are difficult to simulate complex and changeable infusion scenarios, such as the patient's body shaking and environmental changes during mobile infusion, which restricts the cultivation of students' ability to handle actual complex situations. With the continuous development of medical technology and the increasing requirements for the professional qualities of nursing staff, the traditional infusion teaching mode has been difficult to meet the needs. The rise of digital twin technology has brought new hope to infusion teaching. It can achieve real - time interaction and mapping between virtual models and the physical world, and is expected to break through the limitations of traditional teaching and provide innovative and efficient solutions for infusion teaching. Summary of the Invention

[0003] The purpose of the present invention is to provide a mobile infusion teaching experiment system based on digital twin technology, including: an infusion operation module, which includes a real infusion device and a physical simulation scenario. The real infusion device includes an infusion model human, an infusion set, and an infusion needle. The infusion operation module also includes sensors for collecting students' operation data, and the operation data at least includes puncture angle, puncture force, infusion speed, and operation technique.

[0004] A data acquisition and transmission module, connected to the infusion operation module, for real - time transmitting the collected operation data to the digital twin modeling module through a wireless transmission device or a wired manner. The wireless transmission device uses Wi - Fi or Bluetooth technology.

[0005] A digital twin modeling module, for constructing a virtual model corresponding to the physical infusion scenario based on three - dimensional modeling technology, and realizing real - time mapping between the virtual model and the physical operation to dynamically display the operation results. The operation results at least include puncture depth, infusion flow rate, and patient reaction.

[0006] The mobile scenario simulation module, connected to the digital twin modeling module, is used to provide the function of simulating the mobile infusion scenario, including simulating the movement of the patient's body and environmental changes. The movement of the patient's body includes going up and down stairs and wheelchair movement, and the environmental changes include bumping and sudden stops. The mobile scenario simulation module restores complex scenarios through virtual reality or augmented reality technology;

[0007] The real-time feedback module, connected to the digital twin modeling module, is used to analyze the errors in the student's operation based on artificial intelligence algorithms and provide instant feedback. The instant feedback includes visual cues, voice guidance, and tactile simulation. The visual cue is to identify the error area with a specific color, the voice guidance is to play the pre-recorded voice for correcting errors, and the tactile simulation is to provide vibration cues through a vibration device;

[0008] The intelligent evaluation module, connected to the digital twin modeling module, is used to monitor the student's infusion operation in real time and record data, and generate a skill evaluation report. The evaluation indicators of the skill evaluation report at least include puncture accuracy, infusion speed control, patient safety, and operation standardization;

[0009] The teacher management module, connected to the digital twin modeling module, is used for teachers to monitor the student's operation in real time, view the operation effect in the virtual scenario, upload teaching cases, edit simulation scenarios, and formulate evaluation criteria;

[0010] The database module, connected to the digital twin modeling module, intelligent evaluation module, and teacher management module, is used to store student training data, evaluation results, teaching cases, and system logs, and support data mining and analysis to optimize teaching content and provide personalized guidance;

[0011] Among them, the 3D modeling technology in the digital twin modeling module uses the Unity 3D engine, the artificial intelligence algorithm in the real-time feedback module is built based on the deep learning algorithm, and the skill evaluation in the intelligent evaluation module is realized based on big data and machine learning algorithms.

[0012] Furthermore, the multiple sensors include a pressure sensor, a flow sensor, and a motion capture device. The pressure sensor is used to detect the puncture force, the flow sensor is used to monitor the infusion speed, and the motion capture device is used to record the details of the operation actions.

[0013] Furthermore, the artificial intelligence algorithm in the real-time feedback module includes a precise puncture path planning algorithm model, and the precise puncture path planning algorithm model is built based on a convolutional neural network and a recurrent neural network, specifically including:

[0014] Feature extraction is performed on the medical image data of the venous blood vessels of a large number of different patients using a convolutional neural network. The medical image data includes CT scans and ultrasound images. The convolutional neural network includes multiple convolutional layers, pooling layers, and fully connected layers, and batch normalization and Dropout techniques are adopted;

[0015] The extracted features are input into a recurrent neural network. Combining the patient's body posture and infusion site information, the optimal puncture path is predicted. The recurrent neural network adopts a long short-term memory network or a gated recurrent unit structure. The body posture is obtained by an action capture device. The infusion sites include the arm and the back of the hand.

[0016] Furthermore, the training process of the precise puncture path planning algorithm model includes:

[0017] Collect medical image data of no less than 1000 different patients. Each case contains at least 5 images at different angles or positions. Preprocess the image data, including image denoising, contrast enhancement, and normalization;

[0018] Obtain the patient's posture data through an action capture device. After quantifying the posture data, associate it with the medical image data; divide the associated data into a training set, a validation set, and a test set. Use the training set to train the convolutional neural network and the recurrent neural network. Adopt the stochastic gradient descent algorithm to optimize the model parameters. The number of training rounds is no less than 100 rounds until the loss function value of the model on the validation set no longer decreases or reaches the preset minimum loss value.

[0019] Furthermore, the artificial intelligence algorithm in the real-time feedback module includes an infusion risk dynamic assessment algorithm model. The infusion risk dynamic assessment algorithm model adopts an algorithm based on the fusion of Bayesian network and fuzzy logic. Specifically, it includes: collecting the patient's physiological parameters, infusion drug characteristics, and infusion environment factors. The physiological parameters include heart rate, blood pressure, and body temperature, which are collected in real time by connecting to a medical monitoring device. The infusion drug characteristics are retrieved from a drug database. The infusion environment factors are collected by an environmental sensor;

[0020] Use a Bayesian network to establish a probability relationship model between these factors to evaluate the possible infusion risks under different factor combinations. The nodes of the Bayesian network include physiological parameters, drug characteristics, and environmental factors. The causal relationships between the nodes are determined according to medical knowledge and expert experience;

[0021] Introduce fuzzy logic to process the uncertainty of data and perform fuzzy processing on the risk level to accurately describe the risk level. The risk levels include low risk, medium risk, and high risk.

[0022] Furthermore, the training process of the infusion risk dynamic assessment algorithm model includes:

[0023] Collect no less than 500 pieces of historical infusion case data, including patients' physiological parameters, infusion drug characteristics, environmental factors, and actual infusion risk situations that occurred;

[0024] Preprocess the physiological parameter data, remove outliers, and smooth it using the moving average filtering method;

[0025] Determine the fuzzy sets of the input variables and output variables of the fuzzy logic system, as well as the fuzzy rules, based on medical knowledge and expert experience;

[0026] Use the historical infusion case data to train the Bayesian network and the fuzzy logic system. Estimate the parameters of the Bayesian network using the maximum likelihood estimation method. Optimize the performance of the fuzzy logic system by adjusting the weights of the fuzzy rules and the parameters of the fuzzy sets. Finally, fuse the two and calculate the infusion risk level using the weighted average method.

[0027] Furthermore, the specific way for the intelligent evaluation module to quantitatively analyze students' skills is as follows:

[0028] According to the puncture accuracy, calculate the deviation distance between the puncture point and the optimal puncture point. The smaller the deviation distance, the higher the score;

[0029] According to the infusion speed control, calculate the deviation rate between the actual infusion speed and the set infusion speed. The smaller the deviation rate, the higher the score; According to the patient safety, evaluate the potential risks caused to the patient during the operation. The smaller the risk, the higher the score;

[0030] According to the operation standardization, check the correctness and integrity of the operation steps against the standard operation procedure. The higher the correct and complete degree, the higher the score.

[0031] Beneficial effects:

[0032] Algorithm models such as precise puncture path planning are based on the principle of deep learning, deeply analyzing the patient's venous blood vessel conditions and individual differences, and providing students with puncture path guidance accurate to the millimeter level. This enables students to accurately insert the needle during simulated operations, significantly reducing the probability of puncture failure and greatly improving the puncture success rate, making students more confident and assured when facing real patients. The real-time feedback module is like a personal coach, closely monitoring the entire operation process. Once there is the slightest deviation in the student's operation, such as improper infusion speed or incorrect puncture angle, immediate and precise feedback will be given through various methods such as vision, voice, and touch. This all-round feedback mechanism prompts students to quickly correct mistakes, rapidly improve the operation accuracy, and form a standardized and precise operation habit. The infusion risk dynamic assessment algorithm model integrates Bayesian networks and fuzzy logic, comprehensively considering the patient's physiological indicators, drug characteristics, and environmental factors, and accurately predicting infusion risks. Whether it is the potential risk of drug allergy or the early signs of phlebitis, they can be detected and warned in a timely manner, allowing students to make preparations in advance. The mobile scenario simulation module realistically reproduces complex infusion scenarios, such as sudden patient movement or drastic environmental changes. Students repeatedly train in these highly restored scenarios, effectively enhancing their emergency handling capabilities, enabling them to calmly respond when encountering emergencies in actual work and ensuring infusion safety. The intelligent assessment module, based on big data and machine learning, comprehensively records and analyzes students' operation data, generating a detailed and objective skill assessment report from multiple dimensions. Teachers can thereby deeply understand students' learning situations, formulate highly targeted teaching strategies, achieve individualized teaching, and promote the improvement of teaching quality. Students can clearly know their own deficiencies based on the assessment report, clarify the improvement direction, stimulate the motivation for autonomous learning, and form a virtuous learning cycle. The teacher management module facilitates teachers to upload rich teaching cases covering various special situations, expanding students' horizons. Teachers can also freely edit the simulation scenarios and adjust the difficulty to meet the needs of different teaching stages, making the teaching content more rich and diverse. With the help of virtual reality or augmented reality technology, the teaching method becomes vivid and lively, and abstract knowledge becomes intuitive and easy to understand, greatly stimulating students' learning interests, enhancing their learning enthusiasm, and injecting new vitality into teaching. Description of the Drawings

[0033] Figure 1 System principle flowchart; Detailed Implementation Modes

[0034] The following further describes the implementation modes of the present invention in conjunction with the drawings and embodiments. The following embodiments are used to illustrate the present invention, but cannot be used to limit the scope of the present invention.

[0035] Embodiment 1

[0036] Collect a large amount of medical imaging data of the venous blood vessels of different patients from the medical imaging databases of cooperative hospitals, including various types such as CT scans and ultrasound images, covering patients of different ages, genders, and health conditions to ensure the diversity and representativeness of the data. The total amount of collected imaging data is not less than 1000 cases, and each case contains at least 5 images of different angles or parts to comprehensively obtain venous blood vessel information. Preprocess the collected medical imaging data, including operations such as image denoising, contrast enhancement, and normalization. Use the median filtering algorithm to remove the noise in the image and improve the image quality; enhance the contrast between the blood vessels and the surrounding tissues through histogram equalization to facilitate subsequent feature extraction; normalize the image pixel values to a specific range (such as 0-1) to make the data more suitable for the training of the neural network model. In the infusion operation module of the experimental system, use a high-precision motion capture device to obtain the body posture data of the patient (simulated human) in real time, including information such as limb positions and joint angles. At the same time, record the infusion sites selected by the students (such as the arm, the back of the hand, etc.) and associate this information with the corresponding medical imaging data. Quantify the posture data, discretize the continuous data such as joint angles to facilitate the input of subsequent algorithm models. For example, divide the joint angles at a certain interval (such as 10°) to form discrete posture states.

[0037] Construct a CNN architecture that includes multiple convolutional layers, pooling layers, and fully connected layers. The convolutional layers use convolutional kernels of different sizes (such as 3×3, 5×5) to extract blood vessel features of different scales. Set multiple pooling layers (such as the max pooling layer) to reduce the data dimension and computational amount. The fully connected layer is used to integrate the feature information extracted by the previous convolutional layers and pooling layers. Introduce the Batch Normalization technology in the CNN model to accelerate the model training process, improve the stability and generalization ability of the model. At the same time, use the Dropout technology to prevent overfitting, randomly discard some neuron connections, and enhance the robustness of the model. Use the preprocessed medical imaging data to train the CNN model, take the imaging data as the input and the blood vessel features as the output. Use the Stochastic Gradient Descent (SGD) algorithm to optimize the model parameters, set the initial value of the learning rate to 0.01, and gradually reduce the learning rate as the number of training rounds increases (such as reducing by 0.1 times every 10 rounds). During the training process, divide the dataset into a training set, a validation set, and a test set according to the ratio of 70%:20%:10%, and use the validation set to monitor the performance of the model to prevent overfitting. The number of training rounds is not less than 100 rounds until the loss function value of the model on the validation set no longer decreases or reaches the preset minimum loss value.

[0038] Based on the vascular features extracted by CNN, an RNN model is constructed to predict the puncture path. The RNN model adopts the long short-term memory network (LSTM) or gated recurrent unit (GRU) structure to process sequential data (the combined sequence of patient posture and infusion site information and vascular features). Set the number of hidden layers of the RNN model to 50 - 100 and adjust according to the actual training effect. Integrate the vascular features extracted by CNN with information such as patient posture and infusion site to form an input sequence. Take the predicted puncture path as the output of the RNN model, and use the cross-entropy loss function to measure the difference between the predicted path and the actual optimal path (annotated by experienced medical staff). Also use the SGD algorithm to optimize the parameters of the RNN model. The training process is similar to that of the CNN model. After multiple rounds of training (not less than 50 rounds), the model can accurately predict the puncture path.

[0039] Use metrics such as accuracy, mean error, and recall to evaluate the performance of the precise puncture path planning algorithm model. Accuracy is used to measure the proportion of correctly predicted puncture paths by the model; mean error is used to evaluate the average deviation between the predicted path and the actual optimal path, in millimeters; recall is used to evaluate the ability of the model to correctly predict the puncturable path. Calculate these evaluation metrics on the test set to ensure that the model has good performance on unseen data. For example, an accuracy above 90%, a mean error controlled within 1 - 2 millimeters, and a recall above 85% indicate good model performance.

[0040] According to the model evaluation results, if the model performance does not meet the expectations, adopt the following optimization strategies. One is to increase the amount of training data, collect more different types of medical image data and patient posture data to further enrich the learning resources of the model. The second is to adjust the model architecture, such as increasing the number of convolutional layers of CNN or changing the hidden layer structure of RNN to improve the expressive ability of the model. The third is to optimize the model hyperparameters, and find the optimal combination of hyperparameters such as learning rate, batch size, and number of hidden layers through methods such as grid search or random search. The fourth is to adopt the ensemble learning method, weight and fuse multiple trained models to improve the stability and accuracy of the model.

[0041] By connecting with professional medical monitoring devices (such as heart rate monitors, sphygmomanometers, thermometers, etc.), collect the patient's physiological parameter data in real time. Set the data collection frequency to 1 - 5 times per minute to ensure that the changes in the patient's physiological state can be obtained in a timely manner. The collected data includes heart rate range (such as 40 - 180 beats per minute), blood pressure range (such as systolic blood pressure 80 - 200 mmHg, diastolic blood pressure 50 - 120 mmHg), body temperature range (such as 35 - 42 °C), etc.

[0042] Preprocess the collected physiological parameter data to remove outliers (i.e., values that deviate significantly from the normal range due to equipment failures or measurement errors). Use the moving average filtering method to smooth the data and reduce the impact of data fluctuations on the evaluation results.

[0043] Query and obtain the characteristic information of the infusion drug from an authoritative drug database, including drug irritation (divided into three levels: high, medium, and low), concentration range (such as 0.1%-50%), drug type (such as antibiotics, nutrient solutions, etc.). Establish the association rules between drug characteristics and risk assessment. For example, high-irritation drugs may lead to a higher risk of phlebitis at higher concentrations.

[0044] Install environmental sensors (such as temperature sensors, humidity sensors, etc.) in the infusion operation module to collect data on the temperature range (such as 15-35°C) and humidity range (such as 30%-80%) of the infusion environment in real time. Integrate the environmental data with the physiological parameter and drug characteristic data and use them together as input factors for infusion risk assessment.

[0045] Based on medical knowledge and expert experience, determine the structure of the Bayesian network. Take the patient's physiological parameters (heart rate, blood pressure, body temperature), infusion drug characteristics (irritability, concentration, type), and environmental factors (temperature, humidity) as the nodes of the network, and establish the causal relationships between the nodes. For example, abnormal heart rate may be related to drug allergic reactions, and high-concentration irritant drugs may increase the risk of phlebitis in a high-temperature environment. Determine the appropriate probability distribution for each node. For example, the normal distribution is used for continuous physiological parameters (heart rate, blood pressure, body temperature), and the discrete distribution is used for drug characteristics (irritability level, drug type) and environmental factors (temperature range, humidity range), etc. Design a fuzzy logic system to determine the fuzzy sets of the input variables (physiological parameters, drug characteristics, environmental factors) and the output variable (infusion risk level). For example, divide the heart rate into three fuzzy sets: low, normal, and high, divide the temperature into three fuzzy sets: cold, suitable, and hot, and divide the infusion risk level into three fuzzy sets: low risk, medium risk, and high risk. Define fuzzy rules to establish the fuzzy relationship between the input variables and the output variable based on medical knowledge and expert experience. For example, if the heart rate is high, the drug irritability is high, and the environmental temperature is hot, then the infusion risk is high risk. Use fuzzy inference methods (such as Mamdani inference method or Sugeno inference method) to calculate the risk assessment. Collect a large amount of historical infusion case data (not less than 500 cases), including the patient's physiological parameters, infusion drug characteristics, environmental factors, and the actual infusion risk situation (recorded and evaluated by medical staff). Use this data to train the Bayesian network and the fuzzy logic system. When training the Bayesian network, use the maximum likelihood estimation method to estimate the network parameters (conditional probability table). For the fuzzy logic system, optimize the model performance by adjusting the weights of the fuzzy rules and the parameters of the fuzzy sets. Integrate the Bayesian network and the fuzzy logic system, and use the weighted average method or other integration strategies to comprehensively calculate the infusion risk level according to the advantages of the two in different risk assessment scenarios. For example, when the physiological parameters change greatly, give a higher weight to the Bayesian network; when the influence of drug characteristics and environmental factors is more obvious, give a higher weight to the fuzzy logic system.

[0046] Use the Leave-One-Out method or the k-Fold Cross-Validation method (such as k = 10) to verify the infusion risk dynamic assessment algorithm model. Divide the data set into a training set and a validation set (or k subsets), and take each subset as the validation set in turn, and the remaining subsets as the training set, train the model and evaluate the performance on the validation set. Calculate the evaluation indicators such as accuracy, recall rate, and F1 value of the model on the validation set to evaluate the accuracy and reliability of the model for infusion risk assessment. For example, if the accuracy rate reaches more than 85% and the F1 value reaches more than 0.8, it indicates that the model has good performance.

[0047] According to the verification results, if the model performance is not ideal, the following adjustment measures are taken. First, check whether the Bayesian network structure is reasonable, and adjust the causal relationship between nodes according to the actual data and expert opinions. Second, optimize the fuzzy rules and parameters of the fuzzy logic system, add or modify fuzzy rules, adjust the boundaries and membership functions of fuzzy sets, and improve the adaptability of the model. Third, further collect and analyze data, find factors that may affect the model performance, such as data missing or inaccurate under certain special drugs or extreme environmental conditions, supplement and correct the data, and retrain the model.

[0048] Select nursing major students as the experimental subjects, and randomly divide the students into two groups, the experimental group and the control group, with no less than 30 students in each group. Ensure that there are no significant differences between the two groups of students in terms of basic knowledge level, basic operation skills, etc. (through pre-test evaluations, such as theoretical knowledge examinations, basic operation skill assessments, etc.). Provide the same infusion operation hardware environment for the two groups of students, including infusion model humans, infusion sets, needles and other devices of the same model, as well as similar physical simulation scenarios (such as ward simulation environments). Ensure that the system software used by the experimental group and the control group is exactly the same for all functional modules except for the functional modules related to the innovative algorithm model. This can accurately evaluate the impact of the innovative algorithm model on the teaching effect.

[0049] The students in the experimental group use a mobile infusion teaching experiment system that includes an accurate puncture path planning algorithm model and an infusion risk dynamic assessment algorithm model for infusion skill training. During the training process, the students operate according to the puncture path guidance provided by the system, and at the same time the system evaluates the infusion risk in real time and provides feedback. The students in the control group use a traditional mobile infusion teaching experiment system (without the innovative algorithm model) for training, relying only on the general guidance of the teacher and the basic feedback of the system (such as simple error prompts, without accurate puncture path planning and dynamic risk assessment). The training time and training content of the two groups of students are the same, both undergoing a systematic training for a period of (such as 4 - 6 weeks), with no less than (such as 3 - 5 times) training sessions per week, and each training session lasting [Z] hours (such as 1 - 2 hours).

[0050] After the training, a unified infusion skill assessment is conducted on the two groups of students. The assessment content includes puncture accuracy (such as puncture success rate, puncture deviation distance), infusion speed control accuracy (such as the deviation rate between the actual infusion speed and the set speed), risk response ability (such as the proportion of correct handling measures in simulated risk scenarios), operation standardization (such as the correctness and integrity of operation steps), etc. The assessment uses a standardized scoring scale, and is independently scored by multiple experienced teachers (no less than 3), and the average value is taken as the final assessment score of the students.

[0051] Collect the assessment score data of students in the experimental group and the control group, including the scores of various assessment indicators. At the same time, record the operation data of students during the training process (such as the number of puncture attempts, the frequency of adjusting the infusion rate, etc.) to further analyze the learning process and skill improvement of students. Use statistical methods such as independent sample t-test or analysis of variance to compare the differences in assessment scores between students in the experimental group and the control group. Calculate statistical measures such as the mean and standard deviation of the scores of various assessment indicators for the two groups of students to evaluate the significance level of the differences (for example, a p-value less than 0.05 indicates a significant difference). If the scores of students in the experimental group on assessment indicators such as puncture accuracy, infusion rate control accuracy, risk response ability, and operation standardization are significantly higher than those of students in the control group (p-value less than 0.05), it indicates that the precise puncture path planning algorithm model and the infusion risk dynamic assessment algorithm model can effectively improve the infusion skills and risk response ability of students, with a significant synergistic effect. Further analyze the operation data of students to explore the impact of the innovative algorithm model on the learning process of students. For example, if the number of puncture attempts of students in the experimental group is significantly less than that of students in the control group, it shows that the precise puncture path planning algorithm model helps students master puncture skills faster and improve operation efficiency. Based on the experimental results, draw the conclusion that the innovative algorithm model in the present invention has a significant synergistic effect in the mobile infusion teaching experiment system, can effectively improve the teaching quality, and provide a more effective teaching tool for nursing education. It is constructed based on the convolutional neural network (CNN) and the recurrent neural network (RNN) in deep learning algorithms. First, use CNN to extract features from a large amount of medical image data of different patients' venous blood vessels (including CT scans, ultrasound images, etc.), and learn the features such as the morphology, direction, and depth of different blood vessels. Then, input the extracted features into RNN, and combine information such as the patient's body posture (obtained by a motion capture device) and infusion site (such as the arm, back of the hand, etc.) to predict the optimal puncture path. This algorithm model can adapt to different individual differences of patients and provide personalized puncture path guidance for students to improve puncture accuracy. For example, when facing a patient with thinner and more curved blood vessels, it can plan the safest and most accurate puncture path based on the previously learned characteristics of similar blood vessels, reducing the probability of puncture failure. An algorithm based on the fusion of Bayesian network and fuzzy logic is adopted. Collect multi-source data such as the patient's physiological parameters (such as heart rate, blood pressure, body temperature, etc., obtained by connecting with medical monitoring devices), infusion drug characteristics (such as drug irritation, concentration, etc., retrieved from the drug database), and infusion environmental factors (such as temperature, humidity, etc., collected by environmental sensors). Use the Bayesian network to establish a probability relationship model between these factors to evaluate the possible infusion risks under different factor combinations, such as the risk of drug allergy reaction, phlebitis risk, etc. At the same time, introduce fuzzy logic to handle the uncertainty of data and perform fuzzy processing on the risk level to more accurately describe the risk level (such as low risk, medium risk, high risk).This algorithm model can dynamically evaluate the risks during the infusion process in real time, timely remind students and teachers to take corresponding measures, and ensure the safety of patients. For example, when a patient is receiving a high-concentration irritating drug in a high-temperature environment and has a fast heart rate, the algorithm model can accurately evaluate a relatively high risk of phlebitis and prompt to adjust the infusion rate or take corresponding preventive measures. With the virtual infusion scenario constructed by digital twin technology being synchronized with the physical operation in real time, it highly realistically restores the entire infusion process, making students feel as if they are in a real clinical scenario. Students can obtain accurate error prompts immediately during the operation process, significantly improving learning efficiency and correcting operation deviations in a timely manner. Based on big data and innovative artificial intelligence algorithms, it provides objective, quantitative, and accurate evaluation results to comprehensively measure students' skill levels. It supports the simulation of various complex mobile infusion scenarios, effectively enhancing students' response capabilities in actual scenarios and improving professional qualities. The system supports the convenient access of various mobile devices (such as tablets, VR headsets), enabling teaching and training to be carried out anytime and anywhere, breaking through time and space limitations.

[0052] Example 2

[0053] Using the motion capture device equipped in the infusion operation module, comprehensively record the motion data of students during the infusion operation process. These data cover the position information, posture changes, and motion trajectories of body parts such as the arms and hands, and are accurately stored in the form of time series to ensure that the entire infusion operation process can be fully reflected.

[0054] Cleaning: First, conduct meticulous cleaning on the collected original motion data. By setting reasonable thresholds and algorithm rules, remove the noise data (such as abnormal data points caused by slight device jitter) and obvious outliers (such as extreme data that does not conform to the normal human motion logic) in it to ensure the quality and reliability of the data.

[0055] Normalization: Then, perform normalization processing on the cleaned data with different ranges, mapping it to a unified interval (for example, normalizing the position coordinate data to the interval [0,1]) to facilitate the effective processing of the subsequent neural network and avoid affecting the model training effect due to excessive differences in data scales.

[0056] Data annotation: According to the professional standard process of infusion operation, conduct detailed annotation on the motion data that has been cleaned and normalized. Divide the entire infusion operation into multiple key stages, such as puncture preparation, puncture action, adjusting the infusion rate, etc., and accurately label the corresponding tags for the motion data of each stage to provide clear goals for the subsequent supervised learning of the model.

[0057] Input layer: The input layer is carefully designed to receive pre - processed action data. Specifically, the action pose information at each time step is ingeniously transformed into a two - dimensional image, and then a three - dimensional image sequence tensor is formed as the input (dimension: time step × image height × image width). Such an input form can make full use of the advantages of CNN in image data processing and effectively capture the spatial and temporal features in the action data.

[0058] Convolutional layer: Multiple convolutional layers are set up to automatically extract rich features from the action data. In each convolutional layer, several convolutional kernels are equipped. These convolutional kernels perform convolution operations by sliding on the input data and can learn various features of the action from different angles and levels. For example, shallow convolutional kernels may focus on local limb movement details, such as slight changes in the bending angle of the arm and minor pose adjustments of the hand; as the network depth increases, deep convolutional kernels can capture more macroscopic and complex features such as the coordinated movement pattern of the arm and hand during the entire puncture action process and the overall coherence of the action pose.

[0059] Pooling layer: Pooling layers are reasonably added after the convolutional layer, and common max - pooling or average - pooling methods are used to reduce the dimensionality of the extracted features. Through the pooling operation, not only can the data volume be effectively reduced, the computational cost be lowered, and the training and running efficiency of the model be improved, but also key feature information can be retained, preventing the model from overfitting during training and ensuring that the model has good generalization ability.

[0060] Fully - connected layer: After deep feature extraction through multiple convolutional layers and pooling layers, the obtained feature maps are flattened and input into the fully - connected layer. The main role of the fully - connected layer is to comprehensively analyze and classify the various features extracted previously. Based on the input feature vectors, it accurately determines which specific operation stage the student's action trajectory belongs to and whether it strictly conforms to the standard operation process, and outputs corresponding detailed classification results and accurate deviation information, providing strong data support for subsequent real - time feedback and intelligent evaluation.

[0061] Prepare the training dataset: A large amount of pre - processed and labeled action data is scientifically divided into a training set, a validation set, and a test set according to a certain ratio (such as 70% for the training set, 20% for the validation set, and 10% for the test set). The training set serves as the main material for the model to learn, enabling the model to continuously adjust parameters to adapt to the data features; the validation set plays a key role during training. By evaluating the performance of the model on the validation set after each training epoch, various parameters of the model (such as the learning rate, the number of convolutional kernels, etc.) are adjusted in a timely manner to effectively prevent the model from overfitting; the test set is specifically used to objectively and accurately evaluate the final performance of the model after the model training is completed.

[0062] Define the loss function: Considering the characteristics of this classification problem, after careful consideration, the cross-entropy loss function is selected to accurately measure the difference between the model's prediction results and the true labels. For example, when the model predicts whether a student's puncture action meets the standard, the cross-entropy loss function can sensitively capture the degree of prediction error, provide a clear optimization direction for the model's parameter update, and prompt the model to continuously improve the prediction accuracy.

[0063] Select the optimization algorithm: Considering factors such as the model's complexity and training efficiency, the Adam optimization algorithm is finally determined to update the model's parameters. When dealing with the training of deep learning models, the Adam optimization algorithm usually shows excellent convergence speed and good performance, enabling the model to quickly and stably move towards minimizing the loss function during training, effectively improving the training effect of the model.

[0064] Conduct model training: Input the prepared training set data into the carefully constructed CNN model in an orderly manner, and then perform multiple iterative trainings according to the selected loss function and optimization algorithm. In each iteration, the model first carefully calculates the prediction results based on the input data, then accurately calculates the loss value according to the loss function, and finally updates the model's parameters in a timely manner through the optimization algorithm. Repeat this process continuously until the preset training stop conditions are met (such as reaching the specified number of training epochs, the validation set loss no longer decreases, etc.), ensuring that the model can fully learn the internal laws and features in the data and has good performance.

[0065] Use the test set for evaluation: After the model training is completed, use the independently reserved test set to comprehensively and objectively evaluate the performance of the trained model. Commonly used and effective evaluation indicators such as accuracy, recall rate, and F1 value are mainly used to measure the quality of the model. For example, the accuracy rate can intuitively reflect the proportion of correct predictions of the model, that is, the proportion of the number of times the model accurately judges that the student's action trajectory meets or does not meet the standard operation process in the total number of predictions; the recall rate focuses on measuring the ability of the model to correctly predict positive examples (i.e., action trajectories that meet the standard operation); the F1 value is a balanced indicator that comprehensively considers the accuracy rate and the recall rate, and can more comprehensively and objectively evaluate the overall performance of the model. By analyzing these evaluation indicators in detail, the performance of the model in actual applications can be accurately grasped.

[0066] Optimization based on evaluation results: According to the evaluation results of the test set, if it is found that the model performance fails to reach the expected ideal state, it is necessary to specifically optimize and adjust the model. Optimization measures include, but are not limited to, adjusting the model architecture (such as adding or reducing convolutional layers, fine-tuning the convolutional kernel size, etc.), optimizing training parameters (such as appropriately adjusting the learning rate, increasing the amount of training data, etc.) and other methods. Through continuous attempts and adjustments, until the model performance meets the specific requirements of the actual application, ensure that the model can play the best role in the infusion teaching experiment system, providing students with accurate and effective analysis and feedback on the movement trajectory.

[0067] In terms of the accuracy of the puncture action, the average value of the puncture angle deviation of the experimental group students (using the system equipped with the "intelligent analysis algorithm model for infusion movement trajectory") decreased from the initial ±10° to ±3° before and after training, while that of the control group students (using the system with traditional action analysis methods) only decreased from ±10° to ±7°. This shows that the experimental group students have made a more significant improvement in the accuracy of the puncture action, thanks to the "intelligent analysis algorithm model for infusion movement trajectory" being able to analyze the movement trajectory more precisely and timely detect and correct the deviation in the puncture angle.

[0068] Regarding the accuracy of infusion speed control, the proportion of time that the experimental group students could control the infusion speed within the standard range increased from the initial 60% to 90% after training, while that of the control group students increased from 60% to 70%. It can be seen that the algorithm model has played a more obvious promoting role in improving the accuracy of students in this key operation of infusion speed control. Through the comprehensive analysis of the movement trajectory, it can better guide students to adjust the infusion speed operation.

[0069] Statistics show that the average time interval for the experimental group students to receive error prompts during the operation is 3 seconds, while that of the control group students is 8 seconds. This fully demonstrates that the "intelligent analysis algorithm model for infusion movement trajectory" has significant advantages in providing real-time error feedback, being able to more quickly detect errors in students' operations and give prompts in a timely manner, enabling students to correct errors faster and improve learning efficiency.

[0070] In the unified infusion operation assessment after the training course, the average assessment score of the experimental group students is 90 points (out of 100 full marks), and the teachers' subjective evaluations of the operation proficiency and standardization of the experimental group students are generally high, believing that the students perform excellently in aspects such as puncture actions, infusion speed control, and the standardization of the overall operation process. The average assessment score of the control group students is 75 points, and the teachers' evaluations also point out that there are more deficiencies in some operation details of the control group students. Generally speaking, the comprehensive learning effect of the experimental group students is significantly better than that of the control group students, further strongly proving the significant efficiency-enhancing effect of the "intelligent analysis algorithm model for infusion movement trajectory" on infusion teaching.

[0071] Example 3

[0072] Data collection: A large amount of historical student infusion operation data is collected extensively, covering a wide variety of information, including but not limited to operation data (such as precise puncture angle, real-time infusion speed, specific operation time, etc.), patient simulation data (such as the patient's physical condition, movement, etc.), environmental simulation data (such as whether there are bumps, emergency stops, etc.) and corresponding detailed infusion risk event records (such as puncture failure, infusion leakage, patient discomfort, etc.). Ensure that the collected data is sufficiently representative and comprehensive, and can fully reflect the complex relationship between various factors and risk events during the infusion operation.

[0073] Cleaning: The massive amount of data collected is carefully cleaned. Through strict screening and filtering mechanisms, duplicate data (to avoid data redundancy interfering with model training) and obvious outliers (such as extreme data that does not conform to the actual situation) are removed to ensure the purity and reliability of the data.

[0074] Coding: For the classified data (such as the patient's physical status is divided into healthy, mild discomfort, severe discomfort, etc.), reasonable data coding operations are performed to convert it into numerical form for efficient processing of subsequent models. For example, the healthy state is coded as 0, mild discomfort is coded as 1, severe discomfort is coded as 2, etc.

[0075] Normalization: Normalize all collected and organized data of different ranges to the same scale. For example, normalize data such as puncture angle and infusion speed to the interval [0,1] to ensure that different types of data can be processed fairly and effectively in the model, and avoid affecting the model's training effect and prediction accuracy due to large differences in data scales.

[0076] Input layer: The input layer is carefully designed to smoothly receive preprocessed time series data, that is, a combination of operation data, patient simulation data, and environmental simulation data arranged in chronological order. This input form can give full play to the unique advantages of LSTM in processing time series data, allowing the model to accurately capture the association and change trend between data at different times.

[0077] LSTM layer: Set up multiple LSTM layers to deeply process time series data. With its unique internal memory unit and gating mechanism, the LSTM layer can effectively capture the changing patterns of risk factors that may occur in different operation stages, patient conditions, and environmental conditions. For example, it can clearly remember how the infusion risk gradually changes during the puncture process as the puncture angle changes and the patient's body moves, thus providing a solid foundation for subsequent risk prediction.

[0078] Output layer: The output of the LSTM layer is a feature vector that contains a comprehensive description of the current infusion operation status and information on potential risk factors. This feature vector will serve as an important input for the subsequent Bayesian network to further perform accurate prediction of risk probabilities.

[0079] Input layer: The feature vector output by the LSTM layer is used as the input for the Bayesian network, ensuring that the Bayesian network can conduct more in-depth risk analysis and probability prediction based on the information extracted by the LSTM.

[0080] Bayesian network construction: Based on rich historical data and professional domain knowledge, the structure of the Bayesian network is carefully constructed to accurately determine the probability relationships (prior probabilities and conditional probabilities) between each node (representing different risk factors and operation status variables). For example, through the analysis of a large amount of historical data, the conditional probability relationship between the puncture angle and puncture failure, as well as the conditional probability relationship between the patient's body movement status and infusion leakage, etc., are determined, providing a reliable basis for the probability inference of the Bayesian network.

[0081] Probability prediction: Using the powerful probability inference mechanism of the Bayesian network, based on the input feature vector and the established probability relationships, accurate probability predictions are made for various risk events that may occur in the current infusion operation. For example, it can accurately predict the probability of puncture failure under the current puncture action, the probability of infusion leakage under the current patient movement and infusion speed, etc., providing strong guidance for risk prevention and control in the infusion teaching process.

[0082] Prepare the training dataset: The massive data that has been sorted and preprocessed is divided into a training set, a validation set, and a test set according to a scientific and reasonable ratio (such as 70% for the training set, 20% for the validation set, and 10% for the test set). The training set is used to enable the model to continuously learn and adjust parameters to adapt to the internal laws in the data; the validation set plays a key monitoring role during the training process. By evaluating the performance of the model on the validation set after each training epoch, various parameters of the model (such as the number of neurons in the LSTM layer, the structural parameters of the Bayesian network, etc.) are adjusted in a timely manner to effectively prevent the model from overfitting; the test set is used to objectively and accurately evaluate the final performance of the model after the model training is completed.

[0083] Define the loss function: Since the output of the model is the probability prediction value of risk events, after careful consideration, the cross-entropy loss function is selected to accurately measure the difference between the predicted probability and the actual occurrence probability. For example, when predicting the probability of puncture failure, the cross-entropy loss function can sensitively capture the deviation between the predicted value and the actual occurrence of puncture failure, provide a clear optimization direction for the parameter update of the model, and prompt the model to continuously improve the prediction accuracy.

[0084] Select the optimization algorithm: Considering factors such as the complexity of the model and training efficiency, the Adam optimization algorithm is finally determined to update the parameters of the model. The Adam optimization algorithm usually shows excellent convergence speed and good performance when dealing with the training of this hybrid architecture model, enabling the model to quickly and stably move towards minimizing the loss function during the training process, effectively improving the training effect of the model.

[0085] Conduct model training: Input the prepared training set data into the carefully constructed hybrid architecture model in an orderly manner, and then perform multiple iterative trainings according to the selected loss function and optimization algorithm. In each iteration, the model first carefully calculates the prediction result based on the input data, then accurately calculates the loss value according to the loss function, and finally updates the parameters of the model in a timely manner through the optimization algorithm. Repeat this process continuously until the preset training stop condition is reached (such as reaching the specified number of training epochs, the validation set loss no longer decreases, etc.), ensuring that the model can fully learn the internal laws and features in the data and has good performance.

[0086] Use the test set for evaluation: After the model training is completed, use the independently reserved test set to comprehensively and objectively evaluate the performance of the trained model. Mainly focus on the key indicator of prediction accuracy, that is, the proportion of the model correctly predicting whether a risk event occurs. For example, when predicting the probability of puncture failure, if the predicted probability is higher than a certain threshold and the puncture failure actually occurs, or the predicted probability is lower than a certain threshold and the puncture failure does not actually occur, then the prediction is considered correct. Evaluate the accuracy of the model by calculating the proportion of the number of correct predictions to the total number of predictions, so as to accurately grasp the prediction ability of the model in practical applications.

[0087] Optimization based on evaluation results: According to the evaluation results of the test set, if it is found that the performance of the model fails to reach the expected ideal state, it is necessary to targetedly optimize and adjust the model. Optimization measures include, but are not limited to, adjusting the model architecture (such as adding or reducing LSTM layers, fine-tuning the structure of the Bayesian network, etc.), optimizing training parameters (such as appropriately adjusting the learning rate, increasing the amount of training data, etc.), and other methods. Through continuous attempts and adjustments until the performance of the model meets the specific requirements of practical applications, ensure that the model can play the best role in the infusion teaching experiment system, providing students with accurate and effective risk prediction and prevention guidance.

[0088] During the training process, the prediction accuracy of the experimental group of students (using the system equipped with the "dynamic infusion risk prediction algorithm model") for the risk event of puncture failure reached 85%, while that of the control group of students (using the system with traditional risk assessment methods) was only 50%. For the risk event of infusion leakage, the prediction accuracy of the experimental group of students was 80%, and that of the control group of students was 40%. This shows that the "dynamic infusion risk prediction algorithm model" has obvious advantages in risk prediction, being able to more accurately predict various risk events that may occur during the infusion operation process, providing more reliable risk warning information for teachers and students.

[0089] Statistics found that the actual number of puncture failures of the experimental group of students during the training process was 3 times, and the number of infusion leakages was 2 times; while the actual number of puncture failures of the control group of students was 10 times, and the number of infusion leakages was 8 times. This fully shows that the "dynamic infusion risk prediction algorithm model" can more effectively help students and teachers prevent risks in advance. Through timely and accurate risk prediction, students can be more cautious during the operation process, thus significantly reducing the occurrence of risk events.

[0090] In the unified infusion operation assessment after the training course, the average assessment score of the experimental group of students was 92 points (out of 100 full marks), and teachers generally had a high-quality evaluation of the proficiency and standardization of the experimental group of students' operations, believing that the students performed excellently in aspects such as puncture movements, infusion speed control, and the standardization of the overall operation process. The average assessment score of the control group of students was 78 points, and teachers' evaluations also pointed out that there were many deficiencies in some operation details of the control group of students. Generally speaking, the comprehensive learning effect of the experimental group of students was significantly better than that of the control group of students, further strongly proving the significant efficiency-enhancing effect of the "dynamic infusion risk prediction algorithm model" on infusion teaching, because accurate risk prediction can make students more cautious during the operation process, thus improving the operation quality and learning effect.

Claims

1. A mobile infusion teaching experiment system based on digital twin technology, characterized in that, including: An infusion operation module, which includes a real infusion device and a physical simulation scenario. The real infusion device includes an infusion model person, an infusion set, and an infusion needle. The infusion operation module also includes sensors for collecting students' operation data, and the operation data at least includes puncture angle, puncture force, infusion speed, and operation technique; A data collection and transmission module, connected to the infusion operation module, for real-time transmitting the collected operation data to the digital twin modeling module through a wireless transmission device or a wired method. The wireless transmission device uses Wi-Fi or Bluetooth technology; A digital twin modeling module, for constructing a virtual model corresponding to the physical infusion scenario based on 3D modeling technology, and realizing real-time mapping between the virtual model and the physical operation to dynamically display the operation results. The operation results at least include puncture depth, infusion flow rate, and patient reaction; A mobile scenario simulation module, connected to the digital twin modeling module, for providing a mobile infusion scenario simulation function, including simulating patient body movement and environmental changes. The patient body movement includes going up and down stairs and wheelchair movement, and the environmental changes include jolting and sudden stop. The mobile scenario simulation module restores complex scenarios through virtual reality or augmented reality technology; A real-time feedback module, connected to the digital twin modeling module, for analyzing errors in students' operations based on artificial intelligence algorithms and providing instant feedback. The instant feedback includes visual cues, voice guidance, and tactile simulation. The visual cue is to mark the error area with a specific color, the voice guidance is to play a pre-recorded voice for correcting errors, and the tactile simulation is to provide a vibration cue through a vibration device; An intelligent evaluation module, connected to the digital twin modeling module, for real-time monitoring and data recording of students' infusion operations, and generating a skill evaluation report. The evaluation indicators of the skill evaluation report at least include puncture accuracy, infusion speed control, patient safety, and operation standardization; A teacher management module, connected to the digital twin modeling module, for teachers to monitor students' operations in real time, view the operation effects in the virtual scenario, upload teaching cases, edit simulation scenarios, and formulate evaluation criteria; A database module, connected to the digital twin modeling module, the intelligent evaluation module, and the teacher management module, for storing students' training data, evaluation results, teaching cases, and system logs, and supporting data mining analysis to optimize teaching content and provide personalized guidance; Among them, the 3D modeling technology in the digital twin modeling module uses the Unity 3D engine, the artificial intelligence algorithm in the real-time feedback module is constructed based on the deep learning algorithm, and the skill evaluation in the intelligent evaluation module is realized based on big data and machine learning algorithms.

2. The mobile infusion teaching experiment system based on digital twin technology according to claim 1, characterized in that The multiple sensors include a pressure sensor, a flow sensor, and an action capture device. The pressure sensor is used to detect the puncture force, the flow sensor is used to monitor the infusion speed, and the action capture device is used to record the details of the operation actions.

3. The mobile infusion teaching experiment system based on digital twin technology according to claim 1, characterized in that The artificial intelligence algorithm in the real-time feedback module includes an accurate puncture path planning algorithm model, which is constructed based on a convolutional neural network and a recurrent neural network, and specifically includes: Using a convolutional neural network to extract features from a large amount of medical image data of different patients' venous blood vessels. The medical image data includes CT scans and ultrasound images. The convolutional neural network includes multiple convolutional layers, pooling layers, and fully connected layers, and batch normalization and Dropout techniques are adopted; Input the extracted features into a recurrent neural network, and combine the patient's body posture and infusion site information to predict the optimal puncture path. The recurrent neural network adopts a long short-term memory network or a gated recurrent unit structure. The body posture is obtained by an action capture device, and the infusion sites include the arm and the back of the hand.

4. The mobile infusion teaching experiment system based on digital twin technology according to claim 3, wherein The training process of the accurate puncture path planning algorithm model includes: Collect medical image data of no less than 1000 different patients. Each case contains at least 5 images of different angles or parts. Preprocess the image data, including image denoising, contrast enhancement, and normalization; Obtain the patient's posture data through an action capture device, quantify the posture data and associate it with the medical image data; divide the associated data into a training set, a validation set, and a test set. Use the training set to train the convolutional neural network and the recurrent neural network, and adopt the stochastic gradient descent algorithm to optimize the model parameters. The number of training rounds is no less than 100 rounds until the loss function value of the model on the validation set no longer decreases or reaches the preset minimum loss value.

5. The mobile infusion teaching experiment system based on digital twin technology according to claim 1, wherein The artificial intelligence algorithm in the real-time feedback module includes an infusion risk dynamic assessment algorithm model, which adopts an algorithm based on the fusion of a Bayesian network and fuzzy logic, and specifically includes: Collect the patient's physiological parameters, infusion drug characteristics, and infusion environment factors. The physiological parameters include heart rate, blood pressure, and body temperature, which are collected in real time by connecting with medical monitoring devices. The infusion drug characteristics are retrieved from a drug database, and the infusion environment factors are collected by environment sensors; Use a Bayesian network to establish a probability relationship model between these factors to evaluate the possible infusion risks under different factor combinations. The nodes of the Bayesian network include physiological parameters, drug characteristics, and environment factors, and the causal relationships between the nodes are determined according to medical knowledge and expert experience; Introduce fuzzy logic to handle the uncertainty of data and perform fuzzy processing on the risk level to accurately describe the risk level. The risk levels include low risk, medium risk, and high risk.

6. The mobile infusion teaching experiment system based on digital twin technology according to claim 5, wherein The training process of the infusion risk dynamic assessment algorithm model includes: Collect no less than 500 historical infusion case data, including patient physiological parameters, infusion drug characteristics, environment factors, and actual infusion risk situations; Preprocess the physiological parameter data, remove outliers and smooth it using a moving average filtering method; Determine the fuzzy sets of the input variables and output variables of the fuzzy logic system, as well as the fuzzy rules, according to medical knowledge and expert experience; Train the Bayesian network and the fuzzy logic system using historical infusion case data, estimate the parameters of the Bayesian network using the maximum likelihood estimation method, optimize the performance of the fuzzy logic system by adjusting the weights of the fuzzy rules and the parameters of the fuzzy sets, and finally fuse the two and calculate the infusion risk level using the weighted average method.

7. The mobile infusion teaching experiment system based on digital twin technology according to claim 1, characterized in that, The specific way for the intelligent evaluation module to quantitatively analyze students' skills is as follows: According to the puncture accuracy, calculate the deviation distance between the puncture point and the optimal puncture point. The smaller the deviation distance, the higher the score. According to the infusion speed control, calculate the deviation rate between the actual infusion speed and the set infusion speed. The smaller the deviation rate, the higher the score. According to the patient safety, evaluate the potential risks caused to the patient during the operation. The smaller the risk, the higher the score. According to the operation standardization, check the correctness and integrity of the operation steps against the standard operation process. The higher the correct and complete degree, the higher the score.

Images