Burn and plastic department nursing communication follow-up visit system based on artificial intelligence

Abstract

The invention discloses a nursing communication follow-up visit system for the department of burn and plastic surgery based on artificial intelligence, and belongs to the technical field of medical follow-up visit systems. The system comprises a nursing wound acquisition module used for acquiring physiological data of a patient, wound image data, nursing operation records and text record data constructed by subjective feedback of the patient in real time; according to the method, multi-modal data are integrated, and a dynamic patient data framework is constructed; physiological indexes are monitored in real time, and a basis is provided for early infection early warning. A wound area is accurately segmented, and quantitative evaluation of the wound is achieved; the emotion and compliance of the patient are analyzed, the nursing compliance is improved, the infection risk detection rate is increased, and the false alarm rate is reduced; based on a random forest model and a dynamic threshold value, hierarchical risk management is realized, a personalized follow-up visit strategy is constructed, resource configuration is optimized, a feedback driving mechanism is introduced, a closed loop is formed, model dynamic updating is realized, nursing standard upgrading is promoted, and system model prediction precision is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of medical follow-up systems, and in particular to a nursing communication follow-up system for burn and plastic surgery departments based on artificial intelligence. Background Art

[0002] Burn care is a complex surgical injury, and its care relies heavily on the experience of medical staff and the active cooperation of patients. Traditional burn care is hospital-centered, managed through regular follow-up visits, wound dressing changes, and patient self-reporting of symptoms.

[0003] However, this approach has significant drawbacks. Nursing effectiveness assessment relies on patients visiting the hospital for examination, making it difficult to promptly detect signs of wound deterioration or infection. Delayed intervention can lead to irreversible damage and delayed monitoring, especially in remote areas or emergency situations.

[0004] Patients' descriptions of pain and healing progress are easily influenced by personal perception, making it difficult for medical staff to quantify and assess them. Differences in experience can also lead to misdiagnosis and subjective bias in wound status.

[0005] High-risk patients and low-risk patients use the same follow-up frequency, resulting in a waste of medical resources. At the same time, some patients are at risk due to insufficient follow-up, and resource allocation is inefficient. Although early electronic health record (EHR) systems have partially achieved data digitization, they are still limited to information storage and retrieval and lack dynamic analysis and decision support capabilities.

[0006] In view of the above technical defects, a solution is now proposed. Summary of the Invention

[0007] The purpose of the present invention is to provide an artificial intelligence-based nursing communication follow-up system for burn and plastic surgery departments to solve the problems raised.

[0008] To achieve the above objectives, the present invention provides the following technical solution: an artificial intelligence-based nursing communication follow-up system for burn and plastic surgery departments, the system comprising:

[0009] The nursing wound acquisition module is used to acquire the patient's physiological data, wound image data, nursing operation records and text record data constructed by the patient's subjective feedback in real time;

[0010] The data joint analysis module is used to standardize the collected data, extract wound area, inflammation indicators, and pain score characteristics, and predict wound healing progress and complication risks based on machine learning models, and send them to the feedback decision module;

[0011] The communication follow-up module generates a personalized follow-up plan based on the analysis results using a random forest model, sends nursing reminders and rehabilitation guidance to patients through text messages, voice or video communications, and stores the follow-up plan and feeds it back to the decision-making module.

[0012] Furthermore, the data joint analysis module processes the physiological data through a filtering algorithm as follows:

[0013] Physiological data include body temperature T, heart rate H, blood oxygen saturation O and local skin temperature T skin , through wearable sensors and IoT devices, real-time collection is carried out, and the timestamp UTC is marked according to the real-time collection progress, and it is stored in a separate document;

[0014] After summarizing and arranging the collected physiological data, the sliding window filtering algorithm is used to eliminate noise. The calculation formula is:

[0015]

[0016] in, It is represented as the filter output value at time t, N is the window size, x i Represented as raw data points, which can be used as an alternative to T UTC , obtain the pre-stored judgment threshold for body temperature and T UTC When conducting comparative analysis, > judgment threshold, an abnormal body temperature alarm signal is triggered and sent to the feedback decision module.

[0017] Furthermore, the model analysis process of the wound image data of the data joint analysis module is as follows:

[0018] Wound image data is collected via a smartphone or dedicated camera. The improved U-Net model is used for analysis to segment the wound area and calculate the wound area S and redness and swelling index R. The specific formula is:

[0019]

[0020] Where I(p) is the binary mask value of pixel p, R p and G p where R is the red and green channel intensity, respectively, and |P| is the total number of pixels. The pre-stored pixel threshold and inflammation peak are obtained, and the calculated area S is accumulated over all I(p) and compared with the pixel threshold. The quantified wound size is determined based on the comparison size. If R is greater than the inflammation peak, an infection risk signal is generated and sent to the feedback decision module.

[0021] Furthermore, the data joint analysis module performs the following judgment and analysis on the text record data:

[0022] The patient's subjective feedback text is recorded through natural language inquiries and the emotional tendency is analyzed by technology using pre-stored comparison parameters. The BERT model is used to extract the feature vector V, and the emotional score E∈[0,1] is output through the Softmax classifier. The calculation formula is: E=Softmax(W·V+b), where W is the weight matrix and b is the bias term. The pre-stored emotional tendency judgment threshold is obtained and compared with the emotional score E. If W>emotional tendency judgment threshold, a positive feedback signal is generated; if W<emotional tendency judgment threshold, an abnormal feedback signal is generated, and the positive feedback signal and the abnormal feedback signal are sent to the feedback decision module.

[0023] Furthermore, the data joint analysis module uses a random forest model to predict the risk of complications, obtains features including wound area, redness and swelling index, and body temperature deviation ΔT, and brings them into the random forest model to output the risk:

[0024]

[0025] Among them, wi represents the weight of the i-th decision tree, and fi(x) represents the classification result of a single tree for feature x.

[0026] Furthermore, the communication follow-up module obtains the complication risk result, and at the same time obtains the risk follow-up threshold range and compares it with the complication risk result: if the complication risk result R isk If the complication risk result Risk is less than the minimum value of the risk follow-up threshold range, the state is judged to be low risk and a low-risk follow-up signal is generated; if the complication risk result Risk is within the risk follow-up threshold range, the state is judged to be medium risk and a medium-risk follow-up signal is generated; if the complication risk result Risk is greater than the maximum value of the risk follow-up threshold range, the state is judged to be high risk and a high-risk follow-up signal is generated, and the generated class risk follow-up signal is sent to the feedback decision module.

[0027] Furthermore, the feedback decision module obtains the predicted healing time y and the actual healing time through the random forest model. Retrieve the information in the communication module to check whether the patient takes the medicine on time. If yes, mark it as 1, otherwise mark it as 0. Mark the medication result as the compliance label c, and the prediction label stored in advance based on the complication risk results. The model parameters are updated by gradient descent method, and its loss function is:

[0028]

[0029] Among them, α and β represent weight coefficients. The pre-stored clinical feedback record threshold and risk misclassification threshold are retrieved. When the clinical feedback shows a large time prediction deviation, such as MSE > record threshold, a primary treatment plan is generated; when the risk misclassification rate is high, such as CrossWntropy < risk misclassification threshold, a secondary treatment plan is generated.

[0030] The beneficial effects of the present invention are:

[0031] 1. The present invention integrates multimodal data such as wearable devices, image analysis, and natural language processing to construct a dynamic patient data framework. After filtering and denoising, data such as body temperature and heart rate can reflect the patient's metabolic status and stress response in real time, provide a basis for early infection warning, and realize continuous monitoring of patients' physiological indicators; based on the improved U-Net model, the wound area is accurately segmented, and the redness and swelling index is combined to dynamically track the progression of inflammation, replacing the subjectivity of traditional visual assessment to achieve quantitative assessment of patients' wounds; through text emotion scoring, the patient's psychological anxiety and pain tolerance related states are identified, communication strategies are adjusted in a targeted manner, nursing compliance is improved, and targeted patient emotion and compliance analysis is constructed. Based on this, the detection rate of infection risk is improved compared with traditional methods, and the false alarm rate is reduced.

[0032] 2. The present invention realizes hierarchical risk management based on random forest model and dynamic threshold, automatically adjusts the frequency and form of follow-up according to the real-time risk level, such as triggering daily video follow-up for high-risk patients to ensure timely intervention and build personalized follow-up strategies for patients; by accurately distinguishing the risk levels of patients, reducing ineffective follow-up of low-risk patients, allowing medical staff to focus on high-risk cases and achieve efficient resource allocation.

[0033] 3. The present invention introduces a feedback-driven mechanism into the system to form a closed loop of "data collection-analysis-intervention-optimization". The prediction model is regularly optimized according to clinical feedback data to adapt to different populations and injury characteristics and realize dynamic model updates. By analyzing historical nursing effects, the best practices are automatically recommended to promote the upgrading of nursing standards throughout the hospital and improve the iteration of nursing plans. The prediction accuracy of the system model is continuously improved, and the accuracy of the average healing time prediction error is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0035] Figure 1 It is the overall flow chart of the system of the present invention;

[0036] Figure 2 This is a flow chart of data analysis by the data joint analysis module of the present invention. DETAILED DESCRIPTION

[0037] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0038] Example 1: Please refer to Figure 1 - Figure 2 As shown, this embodiment is a nursing communication follow-up system for burn and plastic surgery department based on artificial intelligence. The system includes: a nursing wound acquisition module for acquiring the patient's physiological data, wound image data, nursing operation records and text record data constructed by the patient's subjective feedback in real time.

[0039] The data joint analysis module is used to standardize the collected data, extract wound area, inflammation indicators, and pain score features, and predict wound healing progress and complication risks based on a machine learning model, which is then sent to the feedback decision module. The data joint analysis module processes the physiological data using a filtering algorithm as follows:

[0040] Physiological data include body temperature T, heart rate H, blood oxygen saturation O and local skin temperature T skin , real-time collection through wearable sensors and IoT devices, and time stamping according to the real-time collection progress UTC , and store it in a separate document. It should be noted that the heart rate and blood oxygen data are obtained through the smart bracelet and transmitted to the cloud via Wi-Fi. The sampling rate of the smart bracelet is 10Hz and stored in the time series JSON format; {"timestamp": "2023-10-01T14:30:00", "heart_rate": 75, "SpO2": 98}, according to the timestamp UTC Stored in a separate document for subsequent portable retrieval, the X-axis is time, the Y-axis is body temperature T, heart rate H, blood oxygen saturation O and local skin temperature T skin , build a rectangular coordinate system, different colored lines distinguish indicator types, but are not limited to this, and can be built based on actual conditions;

[0041] After summarizing and arranging the collected physiological data, the sliding window filtering algorithm is used to eliminate noise. The calculation formula is:

[0042]

[0043] in, It is expressed as the filter output value at time t. Its physical meaning is the smoothed estimated value of the parameters in the current window. N is the window size, which is used to eliminate instantaneous fluctuations. i Represented as the original data point, which can be replaced by T UTC 、H UTC , O UTC 、T skin UTC The system collects raw data every 0.1 seconds and calculates each parameter independently. Generate smoothed time series data

[0044] When it is replaced by T UTC , obtain the pre-stored judgment threshold for body temperature and T UTC For comparative analysis, when |T UTC -T UTC-1 |> judgment threshold, triggering an abnormal body temperature alarm signal and sending it to the feedback decision module. The feedback decision module copies and archives the relevant data within the timestamp based on the abnormal body temperature alarm signal, marks the abnormal body temperature document collection, and generates text in the format of "Bed No. XX / Abnormal Body Temperature Alarm / Building Follow-up Plan / Data Waiting for Input / Input Connection" based on the abnormal body temperature alarm signal, which is displayed on the supervisory display screen where the doctor and nurse work or the smart device bound to personal information;

[0045] When it is replaced by H UTC , obtain the pre-stored judgment threshold for heart rate and H UTC For comparative analysis, when |H UTC -H UTC-1 |> judgment threshold, triggering an abnormal heart rate alarm signal and sending it to the feedback decision module. The feedback decision module copies and archives the relevant data within the timestamp based on the abnormal heart rate alarm signal, marks the abnormal heart rate document collection, and generates text in the format of "Bed No. XX / Abnormal Heart Rate Alarm / Building Follow-up Plan / Data Waiting for Input / Input Connection" based on the abnormal heart rate alarm signal, which is displayed on the supervisory display screen where the doctor and nurse work or the smart device bound to personal information;

[0046] When it is replaced by O UTC , obtain the pre-stored judgment threshold value and O UTC For comparative analysis, when UTC -O UTC-1|> judgment threshold, triggering an abnormal blood oxygen saturation alarm signal and sending it to the feedback decision module. The feedback decision module copies and archives the relevant data within the timestamp based on the abnormal blood oxygen saturation alarm signal, marks the abnormal blood oxygen saturation document collection, and generates "Bed No. XX / Abnormal Blood Oxygen Saturation Alarm / Build Follow-up Plan / Data Waiting for Input / Input Connection" format text based on the abnormal blood oxygen saturation alarm signal, which is displayed on the supervisory display screen where the doctor and nurse are located or on the smart device bound to personal information;

[0047] When it is replaced by T skin UTC , obtain the pre-stored judgment threshold value for local skin temperature and T UTC For comparative analysis, when |T skin UTC -T skin UTC-1 |> judgment threshold, then trigger a local skin temperature abnormality alarm signal and send it to the feedback decision module. The feedback decision module copies and archives the relevant data within the timestamp based on the local skin temperature abnormality alarm signal, and marks the local skin temperature abnormality document collection. At the same time, based on the local skin temperature abnormality alarm signal, it generates a text format of "Bed No. XX / Local Skin Temperature Abnormal Alarm / Build Follow-up Plan / Data Waiting for Input / Input Connection" and displays it on the supervisory display screen where the doctor and nurse are located or on the smart device bound to personal information;

[0048] If there are multiple abnormal conditions as mentioned above, the generated alarm signal will be displayed in an up and down gradient on the monitoring display screen where the doctor and nurse are located or on the smart device bound to their personal information. At the same time, intelligent adjustments will be made directly according to the personal follow-up plan of the doctor and nurse, and the text "Bed No. XX / Multiple abnormal joint alarms / Go for follow-up immediately / Data waiting for entry / Entry connection" will be generated and displayed on the smart device of the corresponding doctor and nurse.

[0049] The model analysis process of wound image data in the data joint analysis module is as follows:

[0050] Wound image data is collected via a smartphone or dedicated camera. Patients upload wound photos daily. The server calls the OpenCV library for grayscale conversion and histogram equalization preprocessing. The data is then input into the U-Net model to segment the wound boundary. A heat map displays the changes in the wound area, and an overlaid histogram displays the percentage of area reduction. This is not limited to these methods. Heat maps or overlaid histograms can be constructed based on actual needs. During analysis, an improved U-Net model is used to segment the wound area, calculate the area S, and the redness and swelling index R. The specific formula is:

[0051]

[0052] Where I(p) is the binary mask value of pixel p, R p and G p where |P| is the red and green channel intensity respectively, and the total number of pixels. The pre-stored pixel threshold and inflammation peak are obtained. The calculated area S is accumulated and all I(p) are compared with the pixel threshold. The quantified wound size is determined based on the comparison size.

[0053] R reflects the degree of inflammation. If R is greater than the peak inflammation, an infection risk signal is generated and sent to the feedback decision module. After receiving the infection risk signal, the feedback decision module immediately generates "Bed No. XX / Wound Inflammation Infection Risk / Build Follow-up Plan / Wound Inflammation Follow-up Information Record / Enter Connection" style text and displays it on the supervisory display screen where the doctor and nurse are located or on the smart device bound to personal information.

[0054] The data joint analysis module determines and analyzes text record data as follows:

[0055] The patient's subjective feedback text is recorded through natural language inquiries and the sentiment tendency is analyzed by technology with pre-stored comparison parameters. The patient submits a description of pain through the app, such as "burning sensation relieved". The BERT model extracts and classifies keyword vectors, and the sentiment score is stored in the MySQL database. The BERT model is used to extract the feature vector V and output the sentiment score E∈[0,1] through the Softmax classifier. The calculation formula is:

[0056] E = Softmax(W·V+b)

[0057] Where W is the weight matrix, b is the bias term, and the pre-stored emotional tendency judgment threshold is obtained and compared with the emotional score E:

[0058] If W>emotional tendency judgment threshold, a positive feedback signal is generated and sent to the feedback decision module. After receiving the positive feedback signal, the feedback decision module generates a positive patient emotion reminder icon and displays it on the patient's medical record and doctor / nurse follow-up plan information to assist doctors and nurses in adjusting the consultation according to the emotional state;

[0059] If W is less than the emotional tendency judgment threshold, an abnormal judgment feedback signal is generated and sent to the feedback decision module. After receiving the abnormal judgment feedback signal, the feedback decision module generates a patient emotional abnormality reminder icon and displays it on the patient's medical record and doctor / nurse follow-up plan information to assist doctors and nurses in adjusting the consultation according to the emotional state.

[0060] The data joint analysis module uses the random forest model to predict the risk of complications, and obtains features including wound area S, redness and swelling index R, body temperature deviation △T, body temperature deviation The random forest model is trained using the Scikit-learn library. The model is updated every two weeks based on feedback data and is fed into the random forest model to output the risk level R. isk :

[0061]

[0062] Where wi represents the weight of the i-th decision tree, fi(x) represents the classification result of a single tree for feature x, and the pre-stored high-risk threshold is retrieved for comparison with the classification result:

[0063] If R isk > If the risk threshold is exceeded, a high-risk signal is generated, an alert is automatically sent to the doctor / nurse’s mobile app, and an emergency appointment link is generated. The nursing plan adjustment record is stored on the blockchain to ensure traceability of the operation;

[0064] If R isk < high risk threshold, no signal is generated;

[0065] Model output R isk Relationship with features:

[0066] R isk =0.5f S (S)+0.3·f R (R)+0.2f T (ΔT)

[0067] Among them, f S 、f R and f T is the normalization function.

[0068] Example 2: This example is an artificial intelligence-based nursing communication and follow-up system for burn and plastic surgery departments, including communication;

[0069] The follow-up module generates a personalized follow-up plan based on the analysis results using a random forest model, sends nursing reminders and rehabilitation guidance to patients through text messages, voice or video communication, and stores the follow-up plan in the feedback decision module.

[0070] The communication follow-up module obtains complication risk results and also obtains the risk follow-up threshold range for comparison and analysis with the complication risk results:

[0071] If the complication risk result R isk If the value is less than the minimum value of the risk follow-up threshold range, the patient is judged to be in a low-risk state, and a low-risk follow-up signal is generated. The low-risk follow-up signal is sent to the feedback decision module. After receiving the low-risk follow-up signal, the feedback decision module generates a text message reminder once a week and sends it to the doctor's supervisory display screen or the smart device bound to the personal information;

[0072] If the complication risk result R isk If the patient's risk falls within the risk follow-up threshold, the patient is judged to be in a medium-risk state, and a medium-risk follow-up signal is generated. The medium-risk follow-up signal is sent to the feedback decision module. After receiving the medium-risk follow-up signal, the feedback decision module generates a voice guidance every three days and sends it to the doctor's monitoring display screen or the smart device bound to the personal information.

[0073] If the complication risk result R isk If the value is greater than the maximum value of the risk follow-up threshold range, the state is determined to be high risk, and a high-risk follow-up signal is generated. The high-risk follow-up signal is sent to the feedback decision module. After receiving the high-risk follow-up signal, the feedback decision module generates a daily video follow-up and triggers medical staff intervention, and sends it to the doctor's supervision display screen or the smart device bound to personal information.

[0074] The feedback decision module obtains the predicted healing time y and the actual healing time through the random forest model Actual healing time From the electronic medical record, retrieve the communication module to check whether the patient takes the medicine on time. If yes, mark it as 1, otherwise mark it as 0. Mark the medication result as the compliance label c, and the prediction label stored in advance based on the complication risk result The model parameters are updated by gradient descent method, and its loss function is:

[0075]

[0076] Among them, α is the mean square error weight coefficient, which controls the optimization strength of the model for the accuracy of healing time prediction; β is the cross entropy weight coefficient, which controls the optimization strength of the model for the accuracy of risk classification. The pre-stored clinical feedback record threshold and risk misclassification threshold are retrieved;

[0077] When clinical feedback shows a large deviation in time prediction, such as MSE > recording threshold, a first-level treatment plan is generated, α is increased to strengthen the time regression task, and a reminder text is generated and displayed on the doctor's supervisory display screen or the smart device bound to personal information;

[0078] When the risk misclassification rate is high, such as CrossWntropy < the risk misclassification threshold, a secondary treatment plan is generated to increase β to improve classification performance, and a reminder text is generated and displayed on the doctor's supervisory display screen or the smart device bound to personal information;

[0079] It should be noted that the parameter update process loads the latest 100 patient data y from the database every two weeks. c.

[0080] Combining Example 1 and Example 2, the present invention integrates multimodal data to construct a dynamic patient data framework. On the one hand, it monitors physiological indicators in real time to provide a basis for early infection warning; accurately segments wound areas to achieve quantitative wound assessment; analyzes patient emotions and compliance, improves nursing compliance, increases infection risk detection rate and reduces false alarm rate; on the other hand, based on random forest models and dynamic thresholds, it implements hierarchical risk management, constructs personalized follow-up strategies, and optimizes resource allocation. At the same time, it introduces a feedback-driven mechanism to form a closed loop, realizes dynamic model updates, promotes the upgrading of nursing standards, and improves the prediction accuracy of system models.

[0081] The above content is merely an example and explanation of the structure of the present invention. Those skilled in the art may make various modifications or additions to the described specific embodiments or replace them in a similar manner. As long as they do not deviate from the structure of the invention or exceed the scope defined by the claims, they should all fall within the scope of protection of the present invention.

[0082] Throughout this specification, references to terms such as "one embodiment," "example," or "specific example" indicate that the specific features, structures, materials, or characteristics described in conjunction with that embodiment or example are included in at least one embodiment or example of the present invention. In this specification, schematic representations of these terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

[0083] The preferred embodiments of the present invention disclosed above are intended only to help illustrate the present invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the present invention to specific embodiments. Obviously, many modifications and variations are possible based on the contents of this specification. These embodiments are selected and described in detail in this specification to better explain the principles and practical applications of the present invention, thereby enabling those skilled in the art to better understand and utilize the present invention. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. The artificial intelligence-based nursing communication follow-up system for burn and plastic surgery department is characterized by: The system comprises: The nursing wound acquisition module is used to acquire the patient's physiological data, wound image data, nursing operation records and text record data constructed by the patient's subjective feedback in real time; The data joint analysis module is used to standardize the collected data, extract wound area, inflammation indicators, and pain score characteristics, and predict wound healing progress and complication risks based on machine learning models, and send them to the feedback decision module; The communication follow-up module generates a personalized follow-up plan based on the analysis results using a random forest model, sends nursing reminders and rehabilitation guidance to patients through text messages, voice or video communications, and stores the follow-up plan in the feedback decision module.

2. The artificial intelligence-based nursing communication follow-up system for burn and plastic surgery department according to claim 1 is characterized in that: The data joint analysis module processes the physiological data through a filtering algorithm as follows: Physiological data include body temperature T, heart rate H, blood oxygen saturation O and local skin temperature T skin , through wearable sensors and IoT devices, real-time collection is carried out, and the timestamp UTC is marked according to the real-time collection progress, and it is stored in a separate document; After summarizing and arranging the collected physiological data, the sliding window filtering algorithm is used to eliminate noise. The calculation formula is: in, It is represented as the filter output value at time t, N is the window size, x i Represented as raw data points, which can be used as an alternative to T UTC , obtain the pre-stored judgment threshold for body temperature and T UTC When conducting comparative analysis, An abnormal body temperature alarm signal is triggered and sent to the feedback decision module.

3. The artificial intelligence-based nursing communication follow-up system for burn and plastic surgery department according to claim 2 is characterized in that: The model analysis process of the wound image data in the data joint analysis module is as follows: Wound image data is collected via a smartphone or dedicated camera. The improved U-Net model is used for analysis to segment the wound area and calculate the wound area S and redness and swelling index R. The specific formula is: Where I(p) is the binary mask value of pixel p, R p and G p where R is the red and green channel intensity, respectively, and |P| is the total number of pixels. The pre-stored pixel threshold and inflammation peak are obtained, and the calculated area S is accumulated over all I(p) and compared with the pixel threshold. The quantified wound size is determined based on the comparison size. If R is greater than the inflammation peak, an infection risk signal is generated and sent to the feedback decision module.

4. The artificial intelligence-based nursing communication follow-up system for burn and plastic surgery department according to claim 1 is characterized in that: The data joint analysis module performs the following analysis on the text record data: The patient's subjective feedback text is recorded through natural language inquiries and the emotional tendency is analyzed by technology using pre-stored comparison parameters. The BERT model is used to extract the feature vector V, and the emotional score E∈[0,1] is output through the Softmax classifier. The calculation formula is: E=Softmax(W·V+b), where W is the weight matrix and b is the bias term. The pre-stored emotional tendency judgment threshold is obtained and compared with the emotional score E. If W>the emotional tendency judgment threshold, a positive feedback signal is generated; if W<the emotional tendency judgment threshold, an abnormal feedback signal is generated, and the positive feedback signal and the abnormal feedback signal are sent to the feedback decision module.

5. The artificial intelligence-based nursing communication follow-up system for burn and plastic surgery department according to claim 3 is characterized in that: The data joint analysis module uses the random forest model to predict the risk of complications, obtains features including wound area, redness and swelling index, and body temperature deviation ΔT, and brings them into the random forest model to output the risk fan R isk : Among them, w i Represented as the weight of the i-th decision tree, f i (x) represents the classification result of a single tree for feature x.

6. The artificial intelligence-based nursing communication follow-up system for burn and plastic surgery department according to claim 5 is characterized in that: The communication follow-up module obtains the complication risk result and obtains the risk follow-up threshold range for comparison and analysis with the complication risk result: if the complication risk result R isk If the value is less than the minimum value of the risk follow-up threshold range, the state is determined to be low risk and a low-risk follow-up signal is generated; If the complication risk result R isk If the risk is within the follow-up threshold range, the state is determined to be medium risk and a medium risk follow-up signal is generated; If the complication risk result R isk If the value is greater than the maximum value of the risk follow-up threshold range, the state is determined to be high risk, a high-risk follow-up signal is generated, and the generated class risk follow-up signal is sent to the feedback decision module.

7. The artificial intelligence-based nursing communication follow-up system for burn and plastic surgery department according to claim 1 is characterized in that: The feedback decision module obtains the predicted healing time y and the actual healing time through the random forest model Retrieve the results of whether the patient takes the medicine on time in the communication follow-up module, mark it as 1 if it is, otherwise mark it as 0, mark its medication result as compliance label c, and the prediction label stored in advance based on the complication risk result The model parameters are updated by gradient descent method, and its loss function is: Among them, α and β represent weight coefficients. The pre-stored clinical feedback record threshold and risk misclassification threshold are retrieved. When the clinical feedback shows a large time prediction deviation, such as MSE > record threshold, a primary treatment plan is generated; when the risk misclassification rate is high, such as CrossWntropy < risk misclassification threshold, a secondary treatment plan is generated.

Images