Stress injury visual feedback system

Abstract

The invention relates to the technical field of stress injury visualization, in particular to a stress injury visualization feedback system which collects medical data of a patient through an information acquisition module, the medical data comprises personal information, a blood test report and stress injury conditions, preoperative and intraoperative data are covered, and comprehensive monitoring of the state of the patient is ensured. The risk early warning module generates a standardized vector by analyzing the pressure damage condition. Secondly, mapping the vectors to a three-dimensional matrix through a dimension reduction method, inputting a convolutional neural network model in combination with an operation predicted time length and a blood test report, and outputting a risk point location assessment matrix; based on the evaluation matrix, the system automatically matches corresponding risk prevention measures, maps the risk prevention measures to a digital three-dimensional model of a patient through a three-dimensional model reconstruction module, and visually displays injury and risk point locations. According to the system, comprehensive monitoring and recording of the preoperative, intraoperative and postoperative nursing processes are achieved, and synchronous integrity of pressure injury information is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of pressure injury visualization, and in particular to a pressure injury visualization feedback system. Background Art

[0002] Pressure injuries, also known as pressure sores, are injuries caused by continuous pressure or shear force on the skin and underlying tissues, resulting in local tissue ischemia and necrosis. This type of injury is common in patients who are bedridden or have difficulty moving. Especially during surgery, pressure injuries are more likely to occur because patients remain in a fixed position for a long time. Pressure injuries not only prolong the patient's hospital stay and increase treatment costs, but also have a serious impact on the patient's quality of life.

[0003] Pressure injuries have many impacts on surgery. First, patients who undergo surgery must maintain a fixed position for a long time, especially during some complex and long-term surgeries, which can easily lead to long-term compression of local tissues and cause pressure injuries. This not only increases the difficulty of postoperative care, but may also lead to poor wound healing and increase the risk of infection. Second, once pressure injuries occur, they will prolong the patient's hospital stay, increase medical costs, and have a negative impact on the patient's postoperative recovery.

[0004] The existing handover of pressure injuries mainly relies on the experience and manual recording of nursing staff. However, this manual recording method has significant limitations. First of all, manual recording is not only time-consuming and labor-intensive, but also prone to errors. Especially when handing over between medical staff, such as from preoperative care to intraoperative surgery, and then to postoperative care, the transmission of information often relies on notes or verbal demonstrations. This traditional handover method is prone to inaccurate or forgotten information, increasing the risk of further injury to patients when switching between different care links, and because pressure injuries are not convenient to check and verify at any time, it is easy to lead to information asymmetry, affecting the overall care effect of patients. Summary of the invention

[0005] To solve the above problems, the present invention provides a pressure injury visualization feedback system, which realizes comprehensive monitoring and recording of the preoperative, intraoperative and postoperative care processes by visualizing the pressure injury conditions, and achieves synchronous integrity of the pressure injury information.

[0006] To achieve the above object, the technical solution adopted by the present invention is:

[0007] A pressure injury visualization feedback system, comprising: an information collection module, a risk warning module, a three-dimensional model reconstruction module and a visualization module, wherein the information collection module, the risk warning module, the three-dimensional model reconstruction module and the visualization module are connected in sequence;

[0008] The information acquisition module is used to obtain the patient's medical data, which includes personal information, blood test reports and pressure injury conditions, and the pressure injury conditions include preoperative pressure injury data, intraoperative pressure injury data and postoperative pressure injury data;

[0009] The risk warning module is used to analyze the pressure injury situation based on the Transformer model and output the pressure injury standardized vector, map the pressure injury standardized vector into a three-dimensional matrix, use the three-dimensional matrix, the estimated duration of the surgery and the blood test report as the input of the convolutional neural network model, output the risk point assessment matrix; and output the risk prevention measures based on the risk point assessment matrix matching;

[0010] The three-dimensional model reconstruction module is used to perform the following steps:

[0011] Build a digital 3D model based on the patient’s personal information;

[0012] Based on the risk point assessment matrix, the digital 3D model is proportionally mapped to the damage and risk points;

[0013] Associating risk points in the digital 3D model with corresponding risk prevention measures through links;

[0014] The visual feedback module is used to synchronize the digital three-dimensional model to the doctor side and the nurse side, and allow the doctor side and the nurse side to add, delete, and modify the digital three-dimensional model.

[0015] Furthermore, the personal information includes name, age, height, weight and hospital number.

[0016] Furthermore, the method of analyzing the pressure injury situation based on the Transformer model and outputting the pressure injury standardized vector includes the following steps:

[0017] Clean and format pressure injury situations;

[0018] The Transformer model extracts features of the processed pressure injury and generates a standardized vector of the pressure injury, wherein the standardized vector of the pressure injury includes the location of the pressure injury and the stage of the pressure injury.

[0019] Furthermore, the training steps of the Transformer model include:

[0020] Preprocessing of pressure injury data, including standardization and formatting;

[0021] Dividing the preprocessed data set into a training set, a validation set, and a test set, generating a corresponding label for each sample in the set, wherein the label includes the location and staging information of the pressure injury;

[0022] Define the structure of the Transformer model, including the number of encoder and decoder layers, the number of attention heads, the embedding dimension, and the dimension of the feedforward neural network;

[0023] The training set is input into the model for forward propagation, and the output pressure injury normalization vector is calculated;

[0024] Use the cross entropy loss function to calculate the loss between the model output and the true label;

[0025] The gradient of loss relative to model parameters is calculated through the back-propagation algorithm, and the model parameters are updated using the optimizer;

[0026] Use the validation set to evaluate model performance and tune model hyperparameters;

[0027] The generalization ability of the model is evaluated through the test set.

[0028] Furthermore, mapping the pressure injury normalized vector into a three-dimensional matrix comprises the following steps:

[0029] The pressure injury normalized vector output by the Transformer model is reduced in dimension through principal component analysis to obtain a low-dimensional vector.

[0030] According to the spatial position information of the low-dimensional vector, determine its position index in the three-dimensional matrix;

[0031] Encoding the staging information of the injury according to the low-dimensional vector;

[0032] The three-dimensional matrix is ​​mapped based on the encoding of the position index and the stage information in the three-dimensional matrix.

[0033] Furthermore, the formula of the convolutional neural network model is as follows:

[0034] O = softmax(W o ·flatten(pool(σ(conv(M,W c )+b c )))+W s ·S+W b ·B+b o );

[0035] Among them, O is the output risk point assessment matrix; M is the input three-dimensional matrix; S is the estimated duration of the operation; B is the indicator set of the blood test report; conv is the convolution operation; σ is the activation function; pool represents the pooling operation; flatten is the flattening operation; Wc is the weight matrix of the convolutional layer; b c is the bias of the convolutional layer; W s W is the weight of the estimated duration of surgery; b is the indicator weight set of the blood test report; W o is the weight matrix of the convolution output layer; b o is the bias of the convolution output layer.

[0036] Further, outputting risk prevention measures based on the risk point assessment matrix matching includes: based on the risk position and risk level of the risk point assessment matrix, traversing and matching with a pre-entered measure library to output risk prevention measures.

[0037] Furthermore, constructing the digital three-dimensional model based on the patient's personal information includes: selecting the digital three-dimensional model from a predefined human body model library according to the patient's height and weight.

[0038] Furthermore, the visual feedback module is also used to perform post-operative handover verification and store the digital three-dimensional models of the patient before, during and after surgery.

[0039] The beneficial effect of the present invention is that the present invention collects the patient's medical data, including personal information, blood test reports and pressure injury conditions, through the information collection module. These data include preoperative pressure injury data and intraoperative pressure injury data, ensuring comprehensive monitoring of the patient's status throughout the medical care process. Secondly, the risk warning module uses the Transformer model to intelligently analyze the collected pressure injury conditions (medical staff's transcripts or voice-to-text) to generate standardized vectors. These vectors are mapped to a three-dimensional matrix through a dimensionality reduction method, and then combined with the estimated duration of the operation and the blood test report, they are input into the convolutional neural network model to output a risk point assessment matrix. In this way, the system can accurately identify and evaluate risk points at different locations, avoiding errors and omissions in manual records. Based on the risk point assessment matrix, the system can automatically match corresponding risk prevention measures. These measures are mapped to the patient's digital three-dimensional model through a three-dimensional model reconstruction module, intuitively displaying specific injuries and risk points. Doctors and nurses can view and operate the three-dimensional model in real time through a visual feedback module to ensure accurate transmission and timely updating of information. The system realizes comprehensive monitoring and recording of preoperative, intraoperative and postoperative care processes. By synchronously updating and sharing 3D models, medical staff can collaborate better, reduce errors and omissions in information transmission, and enhance the continuity and effectiveness of patient care. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 It is a structural schematic diagram of a pressure injury visualization feedback system in the present invention.

[0041] Figure 2 It is a flow chart of the execution steps of the three-dimensional model reconstruction module in the present invention.

[0042] Figure 3 It is a flowchart of the steps of mapping the pressure injury normalization vector into a three-dimensional matrix in the present invention. DETAILED DESCRIPTION

[0043] See also Figure 1-3 As shown, the present invention relates to a pressure injury visualization feedback system, comprising: an information acquisition module, a risk warning module, a three-dimensional model reconstruction module and a visualization module, wherein the information acquisition module, the risk warning module, the three-dimensional model reconstruction module and the visualization module are connected in sequence;

[0044] The information acquisition module is used to obtain the patient's medical data, which includes personal information, blood test reports and pressure injury conditions, and the pressure injury conditions include preoperative pressure injury data, intraoperative pressure injury data and postoperative pressure injury data;

[0045] The risk warning module is used to analyze the pressure injury situation based on the Transformer model and output the pressure injury standardized vector, map the pressure injury standardized vector into a three-dimensional matrix, use the three-dimensional matrix, the estimated duration of the surgery and the blood test report as the input of the convolutional neural network model, output the risk point assessment matrix; and output the risk prevention measures based on the risk point assessment matrix matching;

[0046] The three-dimensional model reconstruction module is used to perform the following steps:

[0047] Build a digital 3D model based on the patient’s personal information;

[0048] Based on the risk point assessment matrix, the digital 3D model is proportionally mapped to the damage and risk points;

[0049] Associating risk points in the digital 3D model with corresponding risk prevention measures through links;

[0050] The visual feedback module is used to synchronize the digital three-dimensional model to the doctor side and the nurse side, and allow the doctor side and the nurse side to add, delete, and modify the digital three-dimensional model.

[0051] It should be noted that pressure injury conditions include preoperative, intraoperative and postoperative pressure injury data. Pressure injury data can be obtained by medical staff recording the location and severity of pressure injuries in voice after a physical examination of the patient. These voice records will be converted into text data to ensure detailed monitoring of the patient's status throughout the medical process. Pressure injury data can also be obtained as text data of pressure injuries through typing records.

[0052] The information collection module obtains the patient's comprehensive medical data. The patient's medical data includes personal information, blood test reports, and pressure injury conditions. The data on pressure injury conditions covers multiple stages before, during, and after surgery, ensuring that all aspects of the patient's entire medical process are covered. During the data cleaning process, the system first filters out incomplete or obviously erroneous data entries. Then, the data is standardized, such as converting all numerical data into a unified unit to ensure data consistency. The formatting process includes mapping the text information describing the location of the pressure injury to the predefined body part code, and mapping the text information describing the severity of the injury to the corresponding grading standard. The risk warning module parses the cleaned and formatted pressure injury data based on the Transformer model. Specifically, the Transformer model first extracts features from the processed data to generate standardized vectors of pressure injuries containing injury location and staging information. These vectors are reduced in dimension through principal component analysis (PCA) and converted into low-dimensional representations to form a three-dimensional matrix. Then, the three-dimensional matrix, the estimated duration of the operation, and the blood test report are used as inputs to the convolutional neural network model to output the risk point assessment matrix. The matrix can accurately identify and evaluate risk points at different locations and provide timely warning information. In the 3D model reconstruction module, the system selects the basic model that is closest to the patient's body proportions from the predefined human model library based on the patient's personal information such as height and weight. Then, the basic model is adjusted through operations such as stretching and scaling to accurately match the patient's actual body proportions. Based on the risk point assessment matrix, the information of each risk point is proportionally mapped to the digital 3D model to generate a complete 3D injury and risk point distribution model. The system links the risk points in the 3D model with the corresponding risk prevention measures through links to provide specific protection suggestions. The visual feedback module synchronizes the generated digital 3D model to the doctor and nurse ends, supporting real-time updates and synchronous operations. Medical staff use this module to add, delete and modify the digital 3D model to ensure the dynamic accuracy and operability of the information. Through 3D visualization, medical staff can intuitively view and analyze the patient's pressure injury and risk points, so as to make more accurate nursing decisions. This not only improves nursing efficiency, but also reduces errors and omissions in information transmission, and enhances the continuity and effectiveness of patient care.

[0053] Furthermore, the personal information includes name, age, height, weight and hospital number.

[0054] Furthermore, the method of analyzing the pressure injury situation based on the Transformer model and outputting the pressure injury standardized vector includes the following steps:

[0055] Clean and format pressure injury situations;

[0056] The Transformer model extracts features of the processed pressure injury and generates a standardized vector of the pressure injury, wherein the standardized vector of the pressure injury includes the location of the pressure injury and the stage of the pressure injury.

[0057] In some embodiments, the system first cleans and formats the collected pressure injury data. Assume that during a preoperative examination of a patient, the medical staff recorded the location and severity of the patient's pressure injury by voice. The content of the voice record may include: "There is mild redness and swelling on the left buttocks and moderate ulcer on the right heel." The system uses natural language processing technology to convert the voice into text and further structure it into a standardized data format. The first step in the cleaning process is to remove outliers and noise in the data. For example, if the timestamp of a record does not match the actual examination time, or some terms in the description of the pressure injury are not in the standard vocabulary, the system will automatically identify and mark these abnormal data for medical staff to review or automatically process. Subsequently, the system fills in the missing data to ensure the integrity and continuity of the data. If a record lacks information on the severity, the system may infer and fill it in based on the trend of adjacent records or use a missing value filling algorithm. After completing the data cleaning, the system formats the data. The goal of formatting is to convert the text description into structured data. For example, "mild redness and swelling on the left buttock" will be encoded as the location "LB1" and the severity "Grade 1", while "moderate ulcer on the right heel" will be encoded as the location "RH2" and the severity "Grade 2". This standardized encoding format facilitates subsequent data processing and model training. The processed data is input into the Transformer model for feature extraction. The Transformer model is a deep learning model that is particularly suitable for processing sequence data. In the encoder part of the model, the input data passes through multiple self-attention mechanism layers and feedforward neural network layers to extract features layer by layer. The output of each layer not only contains the information of the current layer, but also combines the global context information, allowing the model to capture complex dependencies and potential patterns. During the feature extraction process, the Transformer model encodes each pressure injury data point to generate a high-dimensional feature vector. These feature vectors contain rich information, including the specific location (such as LB1, RH2) and stage (such as Grade 1, Grade 2) of the pressure injury. For example, for "mild redness and swelling on the left buttock", the feature vector generated by the model may contain the location code "LB1", the severity "Grade 1", and other implicit features.

[0058] Furthermore, the training steps of the Transformer model include:

[0059] Preprocessing of pressure injury data, including standardization and formatting;

[0060] Dividing the preprocessed data set into a training set, a validation set, and a test set, generating a corresponding label for each sample in the set, wherein the label includes the location and staging information of the pressure injury;

[0061] Define the structure of the Transformer model, including the number of encoder and decoder layers, the number of attention heads, the embedding dimension, and the dimension of the feedforward neural network;

[0062] The training set is input into the model for forward propagation, and the output pressure injury normalization vector is calculated;

[0063] Use the cross entropy loss function to calculate the loss between the model output and the true label;

[0064] The gradient of loss relative to model parameters is calculated through the back-propagation algorithm, and the model parameters are updated using the optimizer;

[0065] Use the validation set to evaluate model performance and tune model hyperparameters;

[0066] The generalization ability of the model is evaluated through the test set.

[0067] In some embodiments, first, preprocessing the pressure injury data is a key step in training the Transformer model. Assume that the data collected from multiple patients include the location and staging information of the pressure injury, which exists in text form, with descriptions such as "redness and swelling of the left buttocks, level 1", "right heel ulcer, level 2", etc. The preprocessing process includes standardization and formatting. Standardization ensures that all data have unified dimensions and units, such as mapping all location descriptions to a predefined coding system (such as "left buttocks" is mapped to "LB1"), and converting staging information (such as "level 1") into numerical representation (such as "1"). Formatting converts text data into structured data for easy model processing. Next, the preprocessed data set is divided into a training set, a validation set, and a test set. Assume that the total amount of data is 1,000, which is usually divided in a ratio of 8:1:1, that is, 800 are used for training, 100 are used for validation, and 100 are used for testing. Each sample generates a corresponding label after division, and the label includes the location code and staging information of the pressure injury. For example, the label of the sample "Redness and swelling on the left buttocks, level one" in the training set is "LB1,1". In the model definition stage, the structural parameters of the Transformer model are set. Assume that the encoder and decoder each have 6 layers, each layer contains 8 attention heads, the embedding dimension is set to 512, and the dimension of the feedforward neural network is set to 2048. These parameters define the complexity and capacity of the model, which directly affects the learning and generalization capabilities of the model. The training set is input into the model for forward propagation. During the forward propagation process, the model receives input data, gradually extracts features through calculations at each layer, and finally outputs a standardized vector of pressure injuries. For example, when the data "LB1,1" is input, the model calculates and outputs a high-dimensional vector, which contains the degree of injury and other implicit features at that location. In order to train the model, the loss between the output and the true label needs to be calculated. The cross entropy loss function is used here, which is defined as follows:

[0068]

[0069] Among them, y i is the true label; is the predicted probability output by the model. By calculating the loss, the model can measure the difference between the current prediction and the actual situation.

[0070] Use an optimizer (such as the Adam optimizer) to update the model parameters based on the calculated gradients. The Adam optimizer combines momentum and adaptive learning rate adjustment mechanisms, and the update formula is as follows:

[0071]

[0072] Among them, m t and v tare the first-order and second-order momentum estimates of the gradient, β1 and β2 are the decay rates of the momentum term, η is the learning rate, ∈ is a small amount to prevent division by zero, and θ t is the parameter value of step t.

[0073] During the training process, use the validation set to evaluate model performance and adjust hyperparameters. For example, if you find that the loss on the validation set does not decrease with the training process, you may need to adjust the learning rate or model structure (such as increasing the number of attention heads or layers). Finally, evaluate the generalization ability of the model through the test set. The test set is used to test the performance of the model on unseen data. Evaluation indicators include accuracy, recall, and F1 score. Assume that the accuracy of the model on the test set is 90%, indicating that the model has a strong ability to recognize pressure injuries.

[0074] Furthermore, mapping the pressure injury normalized vector into a three-dimensional matrix comprises the following steps:

[0075] The pressure injury normalized vector output by the Transformer model is reduced in dimension through principal component analysis to obtain a low-dimensional vector.

[0076] According to the spatial position information of the low-dimensional vector, determine its position index in the three-dimensional matrix;

[0077] Encoding the staging information of the injury according to the low-dimensional vector;

[0078] The three-dimensional matrix is ​​mapped based on the encoding of the position index and the stage information in the three-dimensional matrix.

[0079] In some embodiments, first, the system obtains a high-dimensional pressure injury normalized vector output by the Transformer model. Assume that the vector is a high-dimensional vector containing 128 dimensions, and each dimension corresponds to a specific pressure injury feature. These features may include the location of the injury, the severity, and related physiological parameters. In practical applications, high-dimensional data may have redundancy and noise, and direct use may affect the efficiency and accuracy of subsequent processing. In order to reduce the complexity of the data, the system uses principal component analysis (PCA) to reduce the dimension of the high-dimensional vector. PCA is a statistical technique that projects the original high-dimensional data into a low-dimensional space through a linear transformation while retaining the main information and features of the data as much as possible. Specifically, the system first calculates the covariance matrix of the high-dimensional vector, and then performs eigenvalue decomposition on it, selecting the first few largest eigenvalues ​​and their corresponding eigenvectors. These eigenvectors form a new low-dimensional space, and the projection of the original high-dimensional vector in this space is the low-dimensional vector. Assume that after PCA processing, the original 128-dimensional high-dimensional vector is reduced to a 3-dimensional low-dimensional vector. This dimensionality reduction not only reduces the redundancy of the data, but also improves the efficiency of subsequent processing. Next, according to the spatial position information of the low-dimensional vector, its position index in the three-dimensional matrix is ​​determined. Assume that the size of the three-dimensional matrix is ​​1000×1000×1000, and each element represents a specific spatial position. Through linear mapping or other spatial transformation methods, the system converts the value of the low-dimensional vector into a position index in the matrix. After determining the position index, the system encodes it according to the injury staging information of the low-dimensional vector. The injury staging information reflects the severity of the pressure injury, for example, the first level represents a mild injury, the second level represents a moderate injury, and the third level represents a severe injury. Assuming that the staging information contained in the low-dimensional vector is the second level, the system encodes the information into a specific value or symbol so that it can be represented in the three-dimensional matrix. Finally, based on the determined position index in the three-dimensional matrix and the encoding of the staging information, the system maps the three-dimensional matrix. Specifically, the system fills the encoded injury staging information in the corresponding position of the three-dimensional matrix. For example, in the (x=50, y=80, z=67) position of the matrix, fill in the value "2" representing the second-level injury. If there are multiple low-dimensional vectors mapped to the same position, the system can use superposition or maximum value method to ensure the integrity and accuracy of the information.

[0080] Furthermore, the formula of the convolutional neural network model is as follows:

[0081] O = softmax(W o ·flatten(pool(σ(conv(M,W c )+b c )))+W s ·S+W b ·B+b o );

[0082] Among them, O is the output risk point assessment matrix; M is the input three-dimensional matrix; S is the estimated duration of the operation; B is the indicator set of the blood test report; conv is the convolution operation; σ is the activation function; pool represents the pooling operation; flatten is the flattening operation; W c is the weight matrix of the convolutional layer; b c is the bias of the convolutional layer; W s W is the weight of the estimated duration of surgery; b is the indicator weight set of the blood test report; W o is the weight matrix of the convolution output layer; b o is the bias of the convolution output layer.

[0083] In some embodiments, a convolution operation is first applied to the input three-dimensional matrix. The purpose of the convolution operation is to extract local features in the input matrix, and the moving convolution kernel slides on the matrix to perform weighted summation and bias addition on each local area to obtain a new feature matrix. For example, if the size of the input three-dimensional matrix is ​​100×100 and the size of the convolution kernel is 3×3, the convolution operation will extract the features of each 3×3 area and generate a new feature matrix. Next, an activation function, such as ReLU (Rectified Linear Unit), is applied to the result after convolution. The role of the activation function is to introduce nonlinearity so that the model can learn more complex features. By zeroing negative values ​​in the convolution result, the activation function ensures that the model can better capture nonlinear relationships when processing data. Subsequently, a pooling operation is performed on the activated feature matrix. The pooling operation can use maximum pooling or average pooling, the purpose of which is to reduce the dimension of the matrix by downsampling the feature matrix while retaining important feature information. For example, the maximum pooling operation selects the maximum value within each pooling window to generate a reduced version of the feature matrix. After the pooling operation, the pooled feature matrix is ​​flattened into a one-dimensional vector. This step is to convert the three-dimensional feature matrix into a one-dimensional vector so that it can be input into the fully connected layer for further processing. The flattening operation ensures that all the information of the feature matrix can be effectively passed to the fully connected layer. In the fully connected layer, the flattened one-dimensional vector is combined with the indicator set of the estimated duration of the surgery and the blood test report to generate a comprehensive feature vector through weighted summation and bias addition. The role of the fully connected layer is to fuse all the input information and extract the most useful features for the final risk assessment. Finally, the softmax activation function is applied to the output of the fully connected layer to generate the final risk point assessment matrix. The softmax function converts the output into a probability distribution to represent the risk assessment results of each location. In this way, the system is able to generate a risk score for each location, reflecting the possibility of pressure injury at that location.

[0084] Further, outputting risk prevention measures based on the risk point assessment matrix matching includes: based on the risk position and risk level of the risk point assessment matrix, traversing and matching with a pre-entered measure library to output risk prevention measures.

[0085] In some embodiments, first, the system generates a risk point assessment matrix, which contains risk scores for various parts of the patient's body, and each score reflects the possibility and severity of pressure injury in that part. Assume that the size of the assessment matrix is ​​100×100, and each element represents the risk information of a specific location, such as a risk score of "3" indicates moderate risk, and a score of "5" indicates high risk. The system then needs to match the risk location and risk level in the assessment matrix with the pre-entered measure library. The measure library contains preventive measures for different risk levels and locations. For example, for mild risks in the buttocks, it is recommended to use a specific type of cushioning pad; for moderate risks in the feet, it is recommended to use special decompression insoles.

[0086] Furthermore, constructing the digital three-dimensional model based on the patient's personal information includes: selecting the digital three-dimensional model from a predefined human body model library according to the patient's height and weight.

[0087] In some embodiments, first, the system obtains the patient's personal information, including height and weight. This information is the basic data for building a digital three-dimensional model. Assume that a patient's height is 170 cm and his weight is 70 kg. The system records and stores this data through the information acquisition module for subsequent use. The system then accesses a predefined human model library. The model library contains a plurality of human models with different height and weight combinations, each of which is pre-built and stored in the system. In specific operation, the system first selects the closest basic model from the model library according to the patient's height and weight. In this example, since the patient's height is 170 cm and his weight is 70 kg, the system will give priority to the model that best matches the combination. If there is a completely matching model in the model library (i.e., height 170 cm, weight 70 kg), the model is directly selected as the basic model. If the patient's height and weight do not have a completely matching model in the model library, the system will select the closest model and make appropriate adjustments. For example, if the patient is 172 cm tall and weighs 68 kg, and there is no model with this exact combination in the model library, the system may select a model with a height of 170 cm and a weight of 70 kg and adjust it. The adjustment process includes stretching and scaling the selected basic model to make it more closely match the patient's actual body proportions. Specifically, the system calculates the proportions that need to be adjusted based on the differences in the patient's height and weight. For example, if the patient is 2 cm taller and 2 kg lighter than the selected model, the system will stretch the model vertically and scale the model overall accordingly.

[0088] Furthermore, the visual feedback module is also used to perform post-operative handover verification and store the digital three-dimensional models of the patient before, during and after surgery.

[0089] Specifically, first, after acquiring and processing all relevant data of the patient, the system uses the visual feedback module to realize the dynamic display and interactive operation of the digital three-dimensional model. This module not only supports real-time viewing of the patient's pressure injury distribution and risk assessment results, but also provides medical staff with a convenient handover verification function, which can be confirmed by signature, fingerprint and face recognition. In specific operations, post-operative handover verification is a key link. Assuming that a patient has a high risk of pressure injury during surgery, after the operation, the nurse responsible for postoperative care needs to accurately understand the patient's intraoperative condition and risk point distribution. To this end, the visual feedback module displays the digital three-dimensional models before, during and after surgery to the nursing team.

[0090] The above implementation modes are merely descriptions of the preferred implementation modes of the present invention, and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary engineering and technical personnel in the field shall fall within the protection scope determined by the claims of the present invention.

Claims

1. A pressure injury visualization feedback system, characterized in that: include: An information collection module, a risk warning module, a three-dimensional model reconstruction module and a visualization module, wherein the information collection module, the risk warning module, the three-dimensional model reconstruction module and the visualization module are connected in sequence; The information acquisition module is used to obtain the patient's medical data, which includes personal information, blood test reports and pressure injury conditions, and the pressure injury conditions include preoperative pressure injury data, intraoperative pressure injury data and postoperative pressure injury data; The risk warning module is used to analyze the pressure injury situation based on the Transformer model and output the pressure injury standardized vector, map the pressure injury standardized vector into a three-dimensional matrix, use the three-dimensional matrix, the estimated duration of the surgery and the blood test report as the input of the convolutional neural network model, output the risk point assessment matrix; and output the risk prevention measures based on the risk point assessment matrix matching; The three-dimensional model reconstruction module is used to perform the following steps: Build a digital 3D model based on the patient’s personal information; Based on the risk point assessment matrix, the digital 3D model is proportionally mapped to the damage and risk points; Associating risk points in the digital 3D model with corresponding risk prevention measures through links; The visual feedback module is used to synchronize the digital three-dimensional model to the doctor side and the nurse side, and allow the doctor side and the nurse side to add, delete, and modify the digital three-dimensional model.

2. A pressure injury visualization feedback system according to claim 1, characterized in that: The personal information includes name, age, height, weight and hospital number.

3. A pressure injury visualization feedback system according to claim 1, characterized in that: The method of analyzing the pressure injury situation based on the Transformer model and outputting the pressure injury standardized vector comprises the following steps: Clean and format pressure injury situations; The Transformer model extracts features of the processed pressure injury and generates a standardized vector of the pressure injury, wherein the standardized vector of the pressure injury includes the location of the pressure injury and the stage of the pressure injury.

4. A pressure injury visualization feedback system according to claim 3, characterized in that: The training steps of the Transformer model include: Preprocessing of pressure injury data, including standardization and formatting; Dividing the preprocessed data set into a training set, a validation set, and a test set, generating a corresponding label for each sample in the set, wherein the label includes the location and staging information of the pressure injury; Define the structure of the Transformer model, including the number of encoder and decoder layers, the number of attention heads, the embedding dimension, and the dimension of the feedforward neural network; The training set is input into the model for forward propagation, and the output pressure injury normalization vector is calculated; Use the cross entropy loss function to calculate the loss between the model output and the true label; The gradient of loss relative to model parameters is calculated through the back-propagation algorithm, and the model parameters are updated using the optimizer; Use the validation set to evaluate model performance and tune model hyperparameters; The generalization ability of the model is evaluated through the test set.

5. A pressure injury visualization feedback system according to claim 1, characterized in that: Mapping the pressure injury normalized vector into a three-dimensional matrix comprises the following steps: The pressure injury normalized vector output by the Transformer model is reduced in dimension through principal component analysis to obtain a low-dimensional vector. According to the spatial position information of the low-dimensional vector, determine its position index in the three-dimensional matrix; Encoding the staging information of the injury according to the low-dimensional vector; The three-dimensional matrix is ​​mapped based on the encoding of the position index and the stage information in the three-dimensional matrix.

6. A pressure injury visualization feedback system according to claim 1, characterized in that: The formula of the convolutional neural network model is as follows: O=softmax(W o ·flatten(pool(σ(conv(M,W c )+b c )))+W s ·S+W b ·B+b o ); Among them, O is the output risk point assessment matrix; M is the input three-dimensional matrix; S is the estimated duration of the operation; B is the indicator set of the blood test report; conv is the convolution operation; σ is the activation function; pool represents the pooling operation; flatten is the flattening operation; W c is the weight matrix of the convolutional layer; b c is the bias of the convolutional layer; W s W is the weight of the estimated duration of surgery; b is the indicator weight set of the blood test report; W o is the weight matrix of the convolution output layer; b o is the bias of the convolution output layer.

7. A pressure injury visualization feedback system according to claim 1, characterized in that: The outputting of risk prevention measures based on the risk point assessment matrix matching includes: based on the risk position and risk level of the risk point assessment matrix, traversing and matching with a pre-entered measure library to output the risk prevention measures.

8. A pressure injury visualization feedback system according to claim 2, characterized in that: The constructing of the digital three-dimensional model based on the patient's personal information includes: selecting the digital three-dimensional model from a predefined human body model library according to the patient's height and weight.

9. A pressure injury visualization feedback system according to claim 1, characterized in that: The visual feedback module is also used to perform post-operative handover verification and store digital three-dimensional models of the patient before, during and after surgery.

Images