Intelligent nursing system after digestive endoscopy based on deep learning

Abstract

The application relates to the technical field of medical care, in particular to a post-endoscopy intelligent nursing system based on deep learning. The system comprises an endoscopy post-operation data acquisition and processing module, an endoscopy post-operation feature analysis module, an endoscopy post-operation recovery risk assessment module and an endoscopy post-operation recovery risk intelligent nursing module. The system can acquire physiological data of a post-endoscopy patient and endoscopy post-operation gastrointestinal tract state image data, uses a long short-term memory network to acquire long short-time sequence features of post-endoscopy physiological indexes, uses a convolutional neural network to identify and acquire post-endoscopy healing abnormal lesion features, inputs the features into a preset deep neural network model for feature fusion, post-operation recovery risk assessment calculation and post-operation intelligent nursing analysis, generates post-endoscopy patient recovery nursing suggestions, and executes corresponding post-endoscopy risk recovery nursing work. The application can accurately analyze post-operation data of a patient, thereby intelligently guiding nursing decision-making work.

Description

Technical Field

[0001] This invention relates to the field of medical and nursing technology, and in particular to an intelligent nursing system for postoperative digestive endoscopy based on deep learning. Background Technology

[0002] Digestive endoscopy, as a minimally invasive procedure, has been widely used in the analysis and care of gastrointestinal diseases. While this technique offers advantages such as minimal trauma and rapid recovery, the quality of postoperative care directly impacts the patient's recovery process and the occurrence of complications. In recent years, technologies based on artificial intelligence and deep learning have been widely applied in the medical field, achieving significant results, particularly in image recognition and patient monitoring. Deep learning models can extract potential patterns and regularities from large amounts of patient data, providing accurate predictions and decision support. In postoperative care following digestive endoscopy, the introduction of deep learning technology can provide more intelligent and personalized nursing services for postoperative patients. However, current applications of deep learning are mainly concentrated in areas such as image analysis and risk prediction. Intelligent methods for postoperative care have not been fully researched, failing to fully utilize information such as the patient's vital signs, laboratory test results, and postoperative symptoms for comprehensive dynamic monitoring and risk prediction. This makes it difficult to achieve precise and personalized postoperative care and to respond in real-time to the complex changes in patients during postoperative recovery, thereby reducing the efficiency of postoperative care for patients after digestive endoscopy. Summary of the Invention

[0003] Therefore, it is necessary for the present invention to provide a deep learning-based intelligent nursing system for postoperative gastrointestinal endoscopy to solve at least one of the above-mentioned technical problems.

[0004] To achieve the above objectives, a deep learning-based intelligent postoperative care system for digestive endoscopy includes the following modules:

[0005] The post-endoscopic data acquisition and processing module is used to acquire post-endoscopic physiological data and post-endoscopic digestive tract imaging data of patients through the API interface, and to perform post-endoscopic data preprocessing and standardization on the post-endoscopic physiological data and post-endoscopic digestive tract imaging data to obtain post-endoscopic physiological standardized data and post-endoscopic digestive tract standardized imaging data.

[0006] The post-endoscopic feature analysis module is used to extract long and short time series features from the standardized physiological data after digestive endoscopy using a long short-term memory network to obtain the long and short time series features of physiological indicators after endoscopy; and to identify the post-operative healing lesion features from the standard digestive tract imaging data after endoscopy using a convolutional neural network to obtain the abnormal healing lesion features after endoscopy, including the degree of lesion corresponding to post-endoscopic digestive tract bleeding, ulceration and infection.

[0007] The postoperative recovery risk assessment module is used to input the long- and short-term temporal characteristics of postoperative physiological indicators and the characteristics of abnormal lesions after postoperative healing into a preset deep neural network model for feature fusion to generate a fused feature set after digestive endoscopy; the postoperative recovery risk assessment is calculated on the fused feature set after digestive endoscopy to obtain the degree of recovery risk after digestive endoscopy.

[0008] The intelligent nursing module for postoperative recovery risk is used to determine the recovery risk level of patients after digestive endoscopy based on the degree of recovery risk after digestive endoscopy, and to perform intelligent postoperative nursing analysis based on the recovery risk level of patients after digestive endoscopy to generate recovery nursing suggestions for patients after digestive endoscopy, so as to carry out corresponding postoperative risk recovery nursing operations.

[0009] Furthermore, the post-endoscopic data acquisition and processing module includes the following functions:

[0010] By using blockchain-based encryption technology and digital certificate authentication, a secure data collection channel is built between the hospital's internal network and external data requesters. This channel records the corresponding postoperative data interaction logs of patients after digestive endoscopy through the distributed ledger of the blockchain and uses digital certificates to perform identity authentication on external data requesters, thus generating a secure data collection channel for postoperative digestive endoscopy.

[0011] The characteristics of postoperative patient data from digestive endoscopy are obtained, including the storage precision and timestamp format of the patient's physiological data, as well as the resolution and encoding method of the endoscopic image data. The data structure of the hospital information system and the digestive endoscopy image storage system is combined to finely adapt the API interface parameters of the external data request end to generate a set of data request adapted API interface parameters, including data filtering conditions and data format conversion rules.

[0012] Based on the secure data acquisition channel after digestive endoscopy and the API interface parameter set adapted to the data request, the authorized external data request terminal sends acquisition requests to the corresponding hospital information system and digestive endoscopy image storage system in the hospital's internal network to obtain the patient's physiological data and digestive tract status image data after digestive endoscopy.

[0013] Postoperative data preprocessing and standardization were performed on physiological data and gastrointestinal imaging data of patients after digestive endoscopy to obtain standardized physiological data and standardized gastrointestinal imaging data after digestive endoscopy.

[0014] Furthermore, the postoperative data preprocessing and standardization of physiological data and postoperative gastrointestinal imaging data of patients after digestive endoscopy includes:

[0015] Abnormalities were removed and standardized from the physiological data of patients after digestive endoscopy to obtain standardized physiological data after digestive endoscopy.

[0016] Image data enhancement was performed on the digestive tract status images after endoscopy, including rotation, translation and flipping, to obtain enhanced digestive tract images after endoscopy.

[0017] Gray-level histogram equalization was performed on the enhanced gastrointestinal images after endoscopy to obtain contrast-balanced images of the gastrointestinal tract after endoscopy.

[0018] Pixel blurring analysis was performed on the contrast-balanced images of the digestive tract after endoscopy to obtain the pixel blurring values ​​of the digestive tract images after endoscopy; based on the pixel blurring values ​​of the digestive tract images after endoscopy, the corresponding contrast-balanced images of the digestive tract after endoscopy were denoised to obtain denoised images of the digestive tract after endoscopy.

[0019] The denoised digestive tract images after endoscopy were standardized to obtain standard digestive tract images after endoscopy.

[0020] Furthermore, the post-endoscopic feature analysis module includes the following functions:

[0021] After the physiological standardization data of digestive endoscopy, the frequency domain transformation of physiological indicators was performed to obtain the periodic rhythm spectrum of physiological indicators after digestive endoscopy.

[0022] The dynamic trend key node analysis of the periodic rhythm spectrum of physiological indicators after digestive endoscopy was carried out by using a slope change detection and extreme point tracking method. The slope change of each physiological indicator curve was analyzed along the time axis. When the slope change exceeds the preset threshold, it is marked as a potential key node. At the same time, the maximum and minimum points of each physiological indicator curve are tracked to obtain the set of dynamic trend key nodes of physiological indicators after endoscopy.

[0023] Based on the dynamic trend of post-endoscopic physiological indicators, long-scale time windows and short-scale time windows are determined on the time axis within the periodic rhythm spectrum of post-endoscopic physiological indicators between each two adjacent key nodes in the key node set. Based on the long-scale time windows and short-scale time windows, the corresponding post-endoscopic physiological standardized data are divided into long-scale and short-scale window sequences to obtain the long-scale physiological sequence data and the short-scale physiological sequence data of endoscopic surgery.

[0024] Long Short-Term Memory (LSTM) networks were used to extract long and short time-series features from long-term and short-term physiological window sequence data of endoscopic surgery to obtain long and short time-series features of postoperative physiological indicators, including low-frequency trend change features of each physiological indicator under the long time window and high-frequency fluctuation features under the short time window.

[0025] Convolutional neural networks were used to identify postoperative healing lesion features from standard endoscopic gastrointestinal imaging data, resulting in abnormal postoperative healing lesion features, including bleeding points and the degree of lesions corresponding to ulcers in the postoperative gastrointestinal healing process.

[0026] Furthermore, the simultaneous tracking of the maximum and minimum points corresponding to each physiological indicator curve includes:

[0027] The frequency characteristics of curve fluctuations of each physiological index within the periodic rhythm spectrum of postoperative physiological indicators of digestive endoscopy were analyzed to obtain the frequency characteristics of curve fluctuations of each postoperative physiological index.

[0028] Based on the curve fluctuation frequency characteristics of each postoperative physiological indicator, the curve fluctuation pattern region of each physiological indicator within the periodic rhythm spectrum of postoperative physiological indicators is divided to generate the curve fluctuation pattern sub-region corresponding to each postoperative physiological indicator.

[0029] The golden section search method was used to perform extreme value search on the sub-region of the curve fluctuation pattern corresponding to each post-endoscopic physiological index, and the search interval of the extreme value point of the curve fluctuation corresponding to each post-endoscopic physiological index was obtained.

[0030] The parabolic approximation method was used to perform curve local extreme point fitting calculations on the search interval of the extreme point of the curve fluctuation corresponding to each postoperative physiological index, so as to obtain the maximum and minimum points corresponding to each physiological index curve.

[0031] Furthermore, the method of using convolutional neural networks to identify postoperative healing lesion features from standard endoscopic gastrointestinal imaging data includes:

[0032] By obtaining the absorption and scattering characteristics of digestive tract tissues under different wavelengths of light through standard images of the digestive tract after endoscopy, and combining the absorption and scattering characteristics of digestive tract tissues under different wavelengths of light with a preset light scattering model, the corresponding standard images of the digestive tract after endoscopy are analyzed for optical parameters to obtain the absorption coefficient and scattering coefficient at each pixel in the images of the digestive tract after endoscopy.

[0033] Based on the absorption and scattering coefficients of each pixel in the post-endoscopic digestive tract image, the corresponding post-endoscopic digestive tract standard image data are used to identify the image texture fractal structure to generate the fractal feature structure of the post-endoscopic digestive tract image, including the fractal feature structure of mucosal folds and vascular branches; the fractal dimension of the post-endoscopic digestive tract image is obtained according to the fractal feature structure of the post-endoscopic digestive tract image.

[0034] By using the gray-level co-occurrence matrix to perform gray-level texture feature analysis on standard endoscopic digestive tract images, the distribution of gray-level texture features of endoscopic digestive tract images in different directions and between pixel pairs at different distances was obtained.

[0035] Based on the fractal dimension of post-endoscopic digestive tract images and the gray-scale texture feature distribution of post-endoscopic digestive tract images in different directions and between distance pixel pairs, a semantic segmentation network model is used to segment the corresponding post-endoscopic digestive tract standard image data into image anatomical regions, thereby obtaining images of various post-endoscopic digestive tract anatomical structure regions.

[0036] Convolutional neural networks were used to extract postoperative healing lesion features from images of various endoscopic gastrointestinal anatomical structures, obtaining postoperative bleeding points, ulcer size, and ulcer depth changes. Based on the ulcer depth changes, the corresponding ulcer expansion rate was obtained. Then, based on the ulcer size and expansion rate, the lesion severity was calculated using a formula to measure the lesion severity of the corresponding postoperative gastrointestinal images, yielding the severity of the ulcers.

[0037] Furthermore, the formula for calculating the degree of ulcer lesions is as follows:

[0038]

[0039] In the formula, D represents the degree of lesion corresponding to the gastrointestinal ulcer after endoscopy, Ω represents the gastrointestinal imaging area after endoscopy, and A xy R represents the size of the ulcer surface of the digestive tract lesion after endoscopy at location (x,y), λ1 is the ulcer area attenuation factor, and R xy λ1 represents the ulcer expansion rate of the digestive tract lesion at location (x,y) after endoscopy, λ2 is the ulcer expansion rate attenuation factor, α1 is the weighting coefficient of the digestive tract ulcer after endoscopy, n is the number of image features corresponding to the digestive tract image after endoscopy, i is the item index corresponding to the image feature, and f is the ulcer expansion rate at location (x,y). i α1 represents the i-th image feature corresponding to the digestive tract image after endoscopy, α2 is the lesion weighting coefficient of the image feature, and η is the correction coefficient for the degree of lesion.

[0040] Furthermore, the post-endoscopic recovery risk assessment module includes the following functions:

[0041] The mean and variance of the physiological indicators and abnormal lesions after endoscopic surgery were obtained by measuring the long and short time-series characteristics of physiological indicators and the abnormal lesions after endoscopic surgery. Based on the mean and variance of the characteristics, Z-score standardization was performed on the physiological indicators and abnormal lesions after endoscopic surgery to obtain the standard characteristics of physiological indicators and abnormal lesions after endoscopic surgery.

[0042] Mutual information analysis was performed on the various dimensions of the standard features of physiological indicators after endoscopy and the various feature components of the standard features of abnormal lesions after endoscopy to obtain the mutual information between the various physiological dimensions and the various abnormal lesion feature components after endoscopy. Based on the mutual information between the various physiological dimensions and the various abnormal lesion feature components after endoscopy, nonlinear correlation assessment analysis was performed on the various dimensions of the standard features of physiological indicators after endoscopy and the various feature components of abnormal lesions after endoscopy to obtain the nonlinear correlation between the various physiological dimensions and the various abnormal lesion feature components after endoscopy.

[0043] A linear correlation assessment analysis was performed on the various dimensions of the standard features of physiological indicators after endoscopy and the various feature components of the standard features of abnormal lesions after endoscopy, and the linear correlation coefficients between the various physiological dimensions and the feature components of abnormal lesions after endoscopy were obtained.

[0044] The long and short time-series features of physiological indicators after endoscopy and the features of abnormal lesions after endoscopy are input into a preset deep neural network model. The nonlinear correlation and linear correlation coefficient between the physiological features of each dimension after endoscopy and the features of each abnormal lesion are combined to perform feature fusion to generate a fused feature set after digestive endoscopy.

[0045] The postoperative recovery risk assessment formula was used to calculate the postoperative recovery risk of the fusion feature set after digestive endoscopy, and the degree of recovery risk after digestive endoscopy was obtained.

[0046] Furthermore, the specific formula for calculating the postoperative recovery risk assessment is as follows:

[0047]

[0048] In the formula, R represents the risk level of recovery after digestive endoscopy, T represents the postoperative observation time range, t represents the time variable parameter, P(t) represents the time series characteristics of postoperative physiological indicators at time point t, Q(t) represents the time series characteristics of abnormal healing lesions at time point t, β represents the risk weighting coefficient for recovery of abnormal healing lesions, H(t) represents the long-term time series characteristics of postoperative physiological indicators at time point t, γ represents the risk weighting coefficient for recovery of postoperative physiological indicators, ε represents the risk influencing factor for recovery after digestive endoscopy, m represents the total number of individual patient characteristics after digestive endoscopy, and X represents the risk factor. j φ is the specific numerical value corresponding to the individual characteristics of the j-th patient after digestive endoscopy. j ξ represents the weighting coefficient corresponding to the individual characteristics of the j-th patient after digestive endoscopy, and ξ represents the correction coefficient for the degree of recovery risk after digestive endoscopy.

[0049] Furthermore, the postoperative recovery risk intelligent care module includes the following functions:

[0050] Acquire clinical experience after digestive endoscopy and determine a risk grading system for postoperative recovery based on this experience.

[0051] Based on the post-endoscopic recovery risk grading system, the recovery risk level of patients after digestive endoscopy is determined according to the degree of recovery risk after digestive endoscopy, including low recovery risk level, medium recovery risk level and high recovery risk level.

[0052] Postoperative intelligent nursing analysis is performed based on the recovery risk level of patients after digestive endoscopy to generate recovery nursing suggestions for patients after digestive endoscopy, so as to carry out corresponding risk recovery nursing tasks after digestive endoscopy.

[0053] The beneficial effects of this invention are:

[0054] The intelligent nursing system for post-endoscopic gastrointestinal care proposed in this invention, based on deep learning, is composed of a post-endoscopic data acquisition and processing module, a post-endoscopic feature analysis module, a post-endoscopic recovery risk assessment module, and a post-operative recovery risk intelligent nursing module. Compared with existing technologies, the beneficial effects of this application are that by using an API interface to automatically acquire physiological data and digestive tract imaging data of patients after digestive endoscopy, real-time monitoring and assessment of the patient's post-operative condition can be achieved. Physiological data such as heart rate, blood pressure, and body temperature, as well as imaging data such as digestive tract images taken by endoscopy, can help doctors comprehensively understand the patient's post-operative condition. Data preprocessing and standardization are important steps in data analysis, which can eliminate noise and bias from different data sources, making the data have a unified standard format, facilitating subsequent analysis and processing. After standardization, the data will be more comparable, which helps to improve the accuracy and reliability of subsequent analysis results. For example, standardization of physiological data can eliminate the influence of individual differences, allowing models to focus on pathological changes rather than simple physiological fluctuations. Standardization of imaging data can improve image contrast and clarity, making subsequent image recognition more accurate, thereby improving the accuracy of lesion identification and providing a high-quality data foundation for subsequent deep learning models, ensuring the accuracy and reliability of prediction and assessment. Secondly, by using Long Short-Term Memory (LSTM) networks to extract temporal features from standardized physiological data, it is possible to capture the long-term and short-term dependencies of physiological parameter changes, which helps to deeply understand the patient's recovery trend and potential risks. For example, LSTM can learn the changing patterns of multiple physiological indicators such as heart rate, blood pressure, and body temperature at different time points to identify risk warning signals during postoperative recovery, such as the occurrence of acute complications. At the same time, by analyzing the gastrointestinal imaging data after endoscopy, CNN can accurately identify lesion features. CNN extracts detailed information from the images through multiple convolutional layers, which can effectively identify abnormal healing lesions in the gastrointestinal tract, such as areas of bleeding, ulceration, or infection. More importantly, CNN can assess the severity of these lesions and provide more detailed guidance for postoperative recovery. Through in-depth analysis of the two data forms (physiological data and imaging data), a comprehensive assessment of the postoperative recovery status can be achieved, effectively predicting potential complications or risks.Then, by inputting long and short time-series features and abnormal healing lesion features into a deep neural network model for feature fusion, a more comprehensive and complete set of recovery assessment features can be obtained. This fusion not only combines the advantages of physiological and imaging data, but also enables intelligent analysis through deep neural network models to uncover complex relationships between different data. The dataset after feature fusion can provide richer contextual information for recovery risk assessment, helping to reveal the interaction of multiple factors during the patient's recovery process. For example, changes in physiological characteristics are caused by lesions inside the digestive tract, while imaging data can accurately reveal the presence and severity of the lesions. Through calculation by the deep learning model, accurate assessment of recovery risk can be achieved, thus providing a scientific basis for postoperative care. This deep fusion step helps in the personalized assessment of the recovery process of patients after digestive endoscopy, improves the accuracy of postoperative risk prediction, helps doctors identify potentially high-risk patients in advance, and provides targeted interventions. Finally, by assessing a patient's recovery risk level, different levels of nursing intervention can be provided, ensuring the rational allocation of nursing resources and avoiding excessive or insufficient intervention. For example, low-risk patients can receive routine care, while high-risk patients require enhanced monitoring and intervention. Intelligent nursing analysis integrates patients' personalized data to provide targeted nursing suggestions, such as dietary adjustments, medication use, and exercise recommendations, thereby promoting rapid postoperative recovery and reducing the occurrence of complications. Furthermore, intelligent nursing analysis based on recovery risk can help hospitals improve nursing efficiency, optimize nursing processes, reduce unnecessary manual intervention, lower medical costs, and enhance the patient's nursing experience. Through this intelligent nursing analysis, refined and personalized management of the postoperative recovery process can be achieved, ensuring that patients receive the most suitable recovery care plan, effectively improving the quality and effectiveness of postoperative care, and thus significantly enhancing the quality and efficiency of postoperative care. Attached Figure Description

[0055] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0056] Figure 1 This is a schematic diagram of the modules of the intelligent nursing system for postoperative digestive endoscopy based on deep learning according to the present invention;

[0057] Figure 2 for Figure 1 A functional flowchart of the postoperative data acquisition and processing module for intraoperative endoscopy.

[0058] Figure 3 for Figure 1 A schematic diagram of the functional flow of the postoperative feature analysis module for intraoperative endoscopy. Detailed Implementation

[0059] The technical system of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0060] To achieve the above objectives, please refer to Figures 1 to 3 This invention provides a deep learning-based intelligent nursing system for postoperative gastrointestinal endoscopy, the system comprising the following modules:

[0061] The post-endoscopic data acquisition and processing module is used to acquire post-endoscopic physiological data and post-endoscopic digestive tract imaging data of patients through the API interface, and to perform post-endoscopic data preprocessing and standardization on the post-endoscopic physiological data and post-endoscopic digestive tract imaging data to obtain post-endoscopic physiological standardized data and post-endoscopic digestive tract standardized imaging data.

[0062] The post-endoscopic feature analysis module is used to extract long and short time series features from the standardized physiological data after digestive endoscopy using a long short-term memory network to obtain the long and short time series features of physiological indicators after endoscopy; and to identify the post-operative healing lesion features from the standard digestive tract imaging data after endoscopy using a convolutional neural network to obtain the abnormal healing lesion features after endoscopy, including the degree of lesion corresponding to post-endoscopic digestive tract bleeding, ulceration and infection.

[0063] The postoperative recovery risk assessment module is used to input the long- and short-term temporal characteristics of postoperative physiological indicators and the characteristics of abnormal lesions after postoperative healing into a preset deep neural network model for feature fusion to generate a fused feature set after digestive endoscopy; the postoperative recovery risk assessment is calculated on the fused feature set after digestive endoscopy to obtain the degree of recovery risk after digestive endoscopy.

[0064] The intelligent nursing module for postoperative recovery risk is used to determine the recovery risk level of patients after digestive endoscopy based on the degree of recovery risk after digestive endoscopy, and to perform intelligent postoperative nursing analysis based on the recovery risk level of patients after digestive endoscopy to generate recovery nursing suggestions for patients after digestive endoscopy, so as to carry out corresponding postoperative risk recovery nursing operations.

[0065] In the embodiments of this invention, please refer to Figure 1 The diagram shown is a schematic representation of the modules of the intelligent postoperative nursing system for digestive endoscopy based on deep learning according to the present invention. In this example, the intelligent postoperative nursing system for digestive endoscopy based on deep learning includes the following modules:

[0066] S1: Post-endoscopic data acquisition and processing module, used to acquire post-endoscopic physiological data and post-endoscopic digestive tract image data of patients through API interface, and to perform post-endoscopic data preprocessing and standardization on the post-endoscopic physiological data and post-endoscopic digestive tract image data to obtain post-endoscopic physiological standardized data and post-endoscopic digestive tract standardized image data.

[0067] In this embodiment of the invention, physiological data and postoperative digestive tract imaging data of patients after digestive endoscopy are obtained from the hospital's electronic health record system or device interface by using an API interface. The physiological data usually includes vital signs such as the patient's blood pressure, heart rate, body temperature, blood oxygen saturation, and respiratory rate. These data are presented in the form of time series and contain some missing values ​​or abnormal data. Therefore, data cleaning is essential. First, missing values ​​are filled in using interpolation or mean imputation. Outliers are identified and corrected using box plots or standard deviation methods. Then, all physiological data undergoes standardization, scaling the value range of each feature to between 0 and 1 using Z-score standardization. For image data, common deep learning image preprocessing methods are used, such as image resizing, normalization, and grayscale conversion. Images are standardized using pixel mean and standard deviation to ensure pixel values ​​fall between 0 and 1. Before inputting the image data into a convolutional neural network (CNN), appropriate cropping and adjustment are performed to meet the dimensionality requirements of the model input. These preprocessing methods yield standardized postoperative physiological and image data from digestive endoscopy, ultimately resulting in standardized postoperative physiological data and standardized postoperative digestive tract image data.

[0068] S2: Post-endoscopic feature analysis module, which uses a long short-term memory network to extract long and short-term features from the physiologically standardized data after digestive endoscopy to obtain the long and short-term features of physiological indicators after endoscopy; and uses a convolutional neural network to identify the post-endoscopic healing lesion features from the standard imaging data of the digestive tract after endoscopy to obtain the abnormal healing lesion features after endoscopy, including the degree of lesion corresponding to post-endoscopic digestive tract bleeding, ulceration and infection.

[0069] In this embodiment of the invention, standardized postoperative physiological data from digestive endoscopy are processed using a Long Short-Term Memory (LSTM) network to extract temporal features. The LSTM model can capture long-term dependencies and temporal patterns in time-series data, making it suitable for analyzing physiological changes in patients after digestive endoscopy. In model construction, a multi-layer LSTM structure is selected, with 128 units per layer. A bidirectional LSTM structure is used to enhance feature extraction capabilities. The input data is the patient's postoperative physiological data. The network effectively transmits past information to the current moment through the Long Short-Term Memory mechanism, thereby extracting the temporal features of physiological indicators. For postoperative digestive tract imaging data… Convolutional Neural Networks (CNNs) are used for feature recognition. CNNs extract local features from images through several convolutional layers and then downsample them through pooling layers to obtain high-level features of the images. The input of the model is postoperative gastrointestinal imaging data. After processing by convolutional layers, features related to postoperative healing lesions are extracted, including but not limited to the manifestations of abnormal lesions such as bleeding, ulcers, and infections. The model is initialized using a pre-trained VGG16 model and then fine-tuned on postoperative imaging data to make the model more adaptable to specific postoperative imaging features. Finally, the features of abnormal lesions in postoperative healing of endoscopy are obtained, including the degree of lesions corresponding to postoperative gastrointestinal bleeding, ulcers, and infections.

[0070] S3: Post-endoscopic recovery risk assessment module, which is used to input the long and short time-series characteristics of post-endoscopic physiological indicators and the characteristics of abnormal lesions after endoscopic healing into a preset deep neural network model for feature fusion to generate a fused feature set after digestive endoscopy; and to calculate the post-endoscopic recovery risk assessment based on the fused feature set after digestive endoscopy to obtain the degree of post-endoscopic recovery risk.

[0071] In this embodiment of the invention, a fused feature set after digestive endoscopy is obtained by fusing physiological data features and imaging data features. First, the long and short temporal features of the extracted post-endoscopic physiological indicators are merged with the features of abnormal healing lesions after endoscopy. They are then connected together using a fully connected layer (FC layer) to form a fused feature vector. Subsequently, this fused feature is input into a deep neural network model for deeper feature learning and combination. In the design of the deep neural network, a multilayer perceptron (MLP) structure is adopted, containing three hidden layers with 256 neurons in each layer. The activation function is ReLU (Rectified Linear Unit). The output layer uses the sigmoid activation function to map the fused features to the range [0,1], representing the probability of recovery risk. Through this process, the result of the recovery risk assessment after digestive endoscopy is obtained. The degree of recovery risk reflects the potential recovery situation of the patient after surgery. A higher value indicates a greater recovery risk, and a lower value indicates a better recovery situation. Finally, the degree of recovery risk after digestive endoscopy is obtained.

[0072] S4: Postoperative recovery risk intelligent nursing module, which is used to determine the recovery risk level of patients after digestive endoscopy based on the degree of recovery risk after digestive endoscopy, and to perform postoperative intelligent nursing analysis based on the recovery risk level of patients after digestive endoscopy, generate recovery nursing suggestions for patients after digestive endoscopy, and execute corresponding postoperative risk recovery nursing operations.

[0073] In this embodiment of the invention, the recovery risk level of patients after digestive endoscopy is determined by setting specific thresholds based on the recovery risk level calculated in the aforementioned steps. A recovery risk scoring standard is set; for example, if the recovery risk probability is greater than 0.85, it is considered high risk; if the recovery risk probability is between 0.46 and 0.85, it is considered medium risk; and if the recovery risk probability is less than 0.45, it is considered low risk. Based on the patient's recovery risk level, personalized postoperative intelligent nursing analysis is further generated. The intelligent nursing analysis integrates the patient's physiological data, imaging data, and risk level to generate specific nursing recommendations. Nursing recommendations for high-risk patients include more frequent monitoring, adjustment of drug treatment plans, and enhanced postoperative care; medium-risk patients require routine care and regular follow-up examinations; and low-risk patients can receive routine care, avoiding excessive intervention. The patient's recovery nursing recommendations are provided to nursing staff and doctors through the hospital's information system to ensure that postoperative nursing work can be effectively implemented, minimize the probability of postoperative complications, and ensure the smooth recovery of patients after surgery. Finally, recovery nursing recommendations for patients after digestive endoscopy are generated to execute corresponding postoperative risk recovery nursing procedures.

[0074] Furthermore, as an embodiment of the present invention, reference is made to... Figure 2 As shown, Figure 1 A functional flowchart of the post-endoscopic data acquisition and processing module is shown in this embodiment. The post-endoscopic data acquisition and processing module includes the following functions:

[0075] S11: Utilize blockchain-based encryption technology and digital certificate authentication to build a secure data collection channel between the hospital's internal network and external data request terminals. This channel records the corresponding postoperative patient data interaction logs through the distributed ledger of the blockchain and uses digital certificates to perform identity authentication on external data request terminals, thereby generating a secure data collection channel for postoperative digestive endoscopy.

[0076] In this embodiment of the invention, by constructing a secure data acquisition channel between the hospital's internal network and the external data requesting end, it is first necessary to introduce blockchain-based encryption technology and digital certificate authentication mechanisms. The construction of this secure channel relies on the distributed ledger function of the blockchain to record detailed logs of each data interaction. Specific steps include creating a private chain through a blockchain platform (such as Ethereum or Hyperledger), defining interaction behaviors and operation record nodes related to post-digestive endoscopy patient data, and verifying the identity of the external data requesting end through a smart contract each time there is a data request or data transmission, and generating corresponding transaction records. These records will be permanently stored in the blockchain ledger to ensure that data access behavior is tamper-proof. To enhance the security of data transmission, digital certificates are used to authenticate the identity of the external data requesting end. The certificate is generated by a trusted Certificate Authority (CA), and digital signature verification is required for each data interaction. After the data requesting end is authenticated, it can interact with the hospital's internal information system or image storage system through the encrypted channel, thereby ensuring the security and privacy of patient information, and ultimately generating a secure data acquisition channel for post-digestive endoscopy.

[0077] S12: Obtain the characteristics of postoperative patient data after digestive endoscopy, including the field storage precision and timestamp format of patient physiological data, as well as the resolution and encoding method of endoscopic image data. Combine the data structures of the hospital information system and the digestive endoscopy image storage system to finely adapt the API interface parameters of the external data request end to generate a set of data request adapted API interface parameters, including data filtering conditions and data format conversion rules.

[0078] In this embodiment of the invention, the process of acquiring the characteristics of postoperative patient data by digestive endoscopy first requires in-depth analysis of different dimensions of the patient data. For physiological data, the focus is on analyzing the precision of data storage. For example, the precision of storage fields for data such as heart rate and blood pressure should be set to two decimal places, and the timestamp format should conform to ISO standards. The 8601 standard (e.g., "2025-01-06T15:30:00Z") is used. For image data, the resolution and encoding method need to be obtained. Common endoscopic image resolution is 1920x1080, and the encoding method can be H.264 or HEVC. Then, by combining the data structures in the hospital information system (HIS) and the digestive endoscopy image storage system (such as PACS), the interface parameters between the two are analyzed. Through detailed review and adaptation of these parameters, the API interface parameter set required by the external data request end is determined. This interface parameter set includes data filtering conditions (e.g., patient data within a specified date range) and data format conversion rules (e.g., converting H.264 encoded images to JPEG format). In addition, to ensure interface compatibility, necessary data field validation and fault tolerance mechanisms need to be introduced into the API interface to cope with different data formats and possible missing values. Finally, a data request adaptation API interface parameter set is generated.

[0079] S13: Based on the secure data acquisition channel after digestive endoscopy and the API interface parameter set for data request adaptation, and using the authorized external data request terminal, send acquisition requests to the corresponding hospital information system and digestive endoscopy image storage system in the hospital's internal network, and acquire the patient's physiological data and digestive tract status image data after digestive endoscopy.

[0080] In this embodiment of the invention, after constructing a secure data acquisition channel for post-endoscopic data collection and adapting the API interface parameter set, the next step is to initiate a data acquisition request based on an authorized external data requester. This process first requires authentication and authorization of the external requester. After successful authentication, the data acquisition request is sent through a secure API interface. The data request is transmitted to the hospital's internal network via an encrypted blockchain channel. Specific request content includes requests for patient physiological data (such as body temperature, blood pressure, heart rate, etc.) and endoscopic image data (such as images of the digestive tract). The data requester connects to the Hospital Information System (HIS) and the Picture Archiving and Communication System (PACS) and initiates the request according to a pre-set API interface parameter set. During the request process, the data is encrypted and transmitted through the blockchain network to ensure that the data is not leaked or tampered with during transmission. Upon receiving the request, the hospital's system queries the corresponding patient data based on the API request content and returns the corresponding data set. The physiological and image data received by the requester are decrypted to ultimately obtain the patient's physiological data and post-endoscopic digestive tract image data.

[0081] S14: Perform postoperative data preprocessing and standardization on the physiological data and digestive tract imaging data of patients after digestive endoscopy to obtain standardized physiological data and standardized digestive tract imaging data after digestive endoscopy.

[0082] In this embodiment of the invention, patient physiological data and endoscopic image data are converted and cleaned to a unified format to meet the needs of subsequent deep learning algorithm analysis. For physiological data, the data first needs to be time-aligned to ensure that data from different sources have consistent timestamps, and then converted according to standardized rules. For example, heart rate and blood pressure data are unified to be recorded once per minute, with the precision adjusted to two decimal places, and the units of each data field need to be unified (e.g., blood pressure is unified to millimeters of mercury). For image data, the original endoscopic images first need to be quality-screened to remove unqualified images caused by blurry images or unclear acquisition, and then size standardization is performed. The images are adjusted to a fixed resolution (e.g., 1920x1080). To improve the efficiency of image data utilization, the images can be compressed using a compression algorithm (e.g., JPEG2000) and the encoding format can be converted to ensure compatibility. In addition, parameters such as color and contrast of the images also need to be optimized through image processing algorithms to unify the presentation effect of the image data. The processed and standardized physiological data and image data will be stored as a standardized dataset for subsequent deep learning model training and analysis, ensuring the integrity and immutability of the data processing process. Finally, standardized physiological data and standardized digestive tract image data after digestive endoscopy are obtained.

[0083] Furthermore, the postoperative data preprocessing and standardization of physiological data and postoperative gastrointestinal imaging data of patients after digestive endoscopy includes:

[0084] Abnormalities were removed and standardized from the physiological data of patients after digestive endoscopy to obtain standardized physiological data after digestive endoscopy.

[0085] In this embodiment of the invention, the collected physiological data of patients after digestive endoscopy are cleaned. Outliers exist in the data, such as extreme or missing values ​​in physiological data like blood pressure, heart rate, and body temperature. Therefore, outlier detection techniques based on statistical methods, such as Z-score or IQR (interquartile range), are used to remove outliers exceeding the normal range. For missing data, interpolation methods (such as linear interpolation or spline interpolation) are used to fill in the missing parts, ensuring data integrity. Next, the data is processed using standardization methods to unify the dimensions of different physiological parameters. Specifically, the Z-score standardization formula is used: Z = σ / (X - μ), where X is the original data, μ is the mean of the data, and σ is the standard deviation. Standardized data can eliminate dimensional differences between different physiological indicators, ensuring that each physiological data point is within the same order of magnitude, ultimately yielding standardized physiological data after digestive endoscopy.

[0086] Preferably, the postoperative digestive tract image data is enhanced by means of rotation, translation and flipping to obtain postoperative digestive tract enhanced image data.

[0087] In this embodiment of the invention, data augmentation processing is performed on the gastrointestinal imaging data obtained after gastrointestinal endoscopy to increase the diversity of the training set and improve the robustness and accuracy of the subsequent model. Specific operations include: first, rotating the original image with a rotation angle range of ±30 degrees to simulate different perspectives during endoscopy; second, translating the image by moving a certain number of pixels horizontally or vertically to ensure the model can learn information from different locations; and third, performing a mirror flip operation to increase data diversity by horizontally or vertically flipping the image. All augmentation operations are performed during the image preprocessing stage to ensure that the basic structural information of the original image is not altered while increasing the amount of data. The augmented image data provides more samples for subsequent analysis and training, avoiding overfitting, and ultimately yielding enhanced gastrointestinal imaging data after endoscopy.

[0088] Preferably, grayscale histogram equalization is performed on the enhanced digestive tract image data after endoscopy to obtain contrast-balanced digestive tract image data after endoscopy.

[0089] In this embodiment of the invention, grayscale histogram equalization is performed to enhance the contrast and clarity of the digestive tract images after endoscopy. The purpose of grayscale histogram equalization is to improve the contrast of the image by adjusting the distribution of grayscale values, thereby making the details in the digestive tract clearer and facilitating subsequent analysis. In the specific implementation process, the grayscale histogram of the image is first calculated, that is, the frequency of occurrence of each grayscale level pixel in the image is counted. Based on the histogram, the cumulative distribution function (CDF) of each grayscale value is calculated. Through the mapping relationship of CDF, each pixel value in the image is mapped to a new grayscale value, thereby adjusting the brightness and contrast of the image, making the grayscale value distribution of the image more uniform, and the details of the digestive tract are better displayed. The image data after this processing has a higher contrast in visual effect, and finally, the contrast-balanced image data of the digestive tract after endoscopy is obtained.

[0090] Preferably, pixel blur analysis is performed on the post-endoscopic digestive tract contrast-balanced image data to obtain the pixel blur value of the post-endoscopic digestive tract image; based on the pixel blur value of the post-endoscopic digestive tract image, image denoising processing is performed on the corresponding post-endoscopic digestive tract contrast-balanced image data to obtain denoised post-endoscopic digestive tract image data.

[0091] In this embodiment of the invention, pixel blur analysis is performed on the contrast-balanced images of the digestive tract after endoscopy to evaluate image sharpness, and further denoising processing is performed. Pixel blur analysis employs common image sharpness evaluation methods, such as the Laplacian operator or the Sobel gradient operator. Specifically, the second derivative of the image is first calculated using the Laplacian operator to obtain the image's edge information. If the image's edge information is relatively blurry, it indicates poor image sharpness; if the edge information is clear, it indicates high image sharpness. The image quality is evaluated based on the magnitude of the image blur value. For images with high blur values, denoising algorithms are used, such as median filtering and Gaussian filtering. These denoising algorithms can effectively remove random noise from the image, preserve image details, and obtain denoised image data, making the digestive tract image clearer and more accurate, ultimately yielding denoised digestive tract image data after endoscopy.

[0092] Preferably, the denoised digestive tract image data after endoscopy is standardized to obtain standard digestive tract image data after endoscopy.

[0093] In this embodiment of the invention, the denoised post-endoscopic gastrointestinal image data is standardized to ensure data consistency and comparability. The standardization method adopts a zero-mean, unit-variance standardization method. The specific steps are as follows: First, the mean and standard deviation of the pixel values ​​of the denoised image are calculated. Then, each pixel value is standardized using the following formula: Q = (xa) / b, where x is the pixel value in the denoised image data, a is the mean of the image pixels, and b is the standard deviation of the image pixels. The standardized image data has a uniform scale in the range of pixel values, which can eliminate the differences caused by the different brightness and contrast of the image itself. This standardization step not only ensures the consistency of the data, but also avoids the adverse effects of brightness differences in the image data on subsequent analysis and deep learning model training, and finally obtains standard post-endoscopic gastrointestinal image data.

[0094] Furthermore, as an embodiment of the present invention, reference is made to... Figure 3 As shown, Figure 1 A functional flowchart of the post-endoscopic feature analysis module is shown in this embodiment. The post-endoscopic feature analysis module includes the following functions:

[0095] S21: Perform frequency domain transformation on the physiological standardized data after digestive endoscopy to obtain the periodic rhythm spectrum of physiological indicators after digestive endoscopy.

[0096] In this embodiment of the invention, frequency domain analysis is performed on the standardized physiological data after digestive endoscopy. This standardized data includes physiological indicators such as the patient's heart rate, respiratory rate, and blood oxygen saturation. After equalization processing, noise and outliers are removed to ensure the accuracy of the data. Then, the Fast Fourier Transform (FFT) algorithm is used to convert these time-domain data to the frequency domain. FFT can efficiently convert signals from the time domain to the frequency domain and generate periodic information spectrum. For each physiological indicator, based on its frequency response characteristics, the periodic rhythm spectrum is obtained by analyzing its frequency domain characteristics. This reveals the periodic pattern of physiological indicators changing over time, and finally, the periodic rhythm spectrum of physiological indicators after digestive endoscopy is obtained.

[0097] S22: The dynamic trend key node analysis of the periodic rhythm spectrum of physiological indicators after digestive endoscopy is carried out by using the slope change detection and extreme point tracking method. The slope change of each physiological indicator curve is analyzed along the time axis. When the slope change exceeds the preset threshold, it is marked as a potential key node. At the same time, the maximum and minimum points corresponding to each physiological indicator curve are tracked to obtain the set of dynamic trend key nodes of physiological indicators after endoscopy.

[0098] In this embodiment of the invention, by performing key node analysis on the dynamic trend of the periodic rhythm spectrum of physiological indicators after digestive endoscopy, firstly, a slope change detection method is used to capture the abrupt change points of each physiological indicator. Specifically, the derivative of the spectrum data is calculated and its change trend is analyzed. If the slope changes abruptly and the change amplitude exceeds a preset threshold, it is considered a potential key node. In this way, the time points when the physiological indicators change drastically can be accurately captured. In addition, the extreme point tracking algorithm is used to locate the maximum and minimum values ​​in the periodic rhythm spectrum. Extreme points usually represent significant fluctuations or stable states of physiological indicators. Combining these slope abrupt change points and extreme points, a set of key nodes can be comprehensively identified. This step effectively identifies potential physiological abnormal moments, ensuring timely monitoring of the patient's postoperative physiological state, and finally obtaining a set of key nodes for the dynamic trend of physiological indicators after endoscopy.

[0099] S23: Based on the dynamic trend of post-endoscopic physiological indicators, determine the long-scale time window and the short-scale time window on the time axis of the periodic rhythm spectrum of post-endoscopic physiological indicators between every two adjacent key nodes in the key node set. Based on the long-scale time window and the short-scale time window, divide the corresponding post-endoscopic physiological standardized data into long-scale and short-scale window sequences to obtain the long-scale physiological sequence data and the short-scale physiological sequence data of endoscopic surgery.

[0100] In this embodiment of the invention, a time window is divided based on the set of key nodes of dynamic trends obtained in the previous step. For each pair of adjacent key nodes, the time interval between them is first calculated on the time axis of the periodic rhythm spectrum. The length of this time interval determines the defined long-scale time window and short-scale time window. The long-scale time window usually covers a longer period of time and is mainly used to analyze the overall trend and chronic fluctuations of physiological indicators. The short-scale time window is used to capture short-term, rapidly changing physiological fluctuations. In specific implementation, the time periods of these two scales are extracted by the sliding window method and the data is divided to ensure that the long-term and short-term physiological dynamic changes can be comprehensively analyzed. This process provides effective input sequence data for subsequent deep learning analysis, and finally, the long-term physiological sequence data and the short-term physiological sequence data of endoscopic surgery are obtained.

[0101] S24: Use a long short-term memory network to extract long and short time series features from the long-term and short-term physiological window sequence data of endoscopic surgery to obtain the long and short time series features of postoperative physiological indicators, including the low-frequency trend change features of each postoperative physiological indicator under the long time window and the high-frequency fluctuation features under the short time window.

[0102] In this embodiment of the invention, a Long Short-Term Memory (LSTM) network is used to extract features from previously segmented long-term and short-term window sequence data. LSTM is a recurrent neural network suitable for time series data, which can effectively learn the long-term and short-term dependencies in time series data. Long-term window sequence data mainly reflects the long-term trend of changes in patients' physiological indicators, while short-term window sequence data contains drastic fluctuations in the short term. The LSTM network processes these two types of data separately and extracts their temporal features. Under the long-term window, LSTM extracts the low-frequency trend change features of physiological indicators, which are usually related to chronic changes and stable states. Under the short-term window, LSTM extracts the high-frequency fluctuation features, which are usually related to acute changes and rapid fluctuations. In this way, the macroscopic trend and microscopic fluctuations of patients' postoperative physiological indicators can be captured simultaneously, and finally, the long and short-term temporal features of postoperative endoscopic physiological indicators are obtained, including the low-frequency trend change features of each postoperative physiological indicator under the long-term window and the high-frequency fluctuation features under the short-term window.

[0103] S25: Using a convolutional neural network, postoperative healing lesion features are identified from standard endoscopic images of the digestive tract to obtain abnormal healing lesion features, including bleeding points and the degree of lesions corresponding to ulcers in the digestive tract after endoscopy.

[0104] In this embodiment of the invention, a convolutional neural network (CNN) is used to identify healing lesion features from standard imaging data of the digestive tract after endoscopy. CNNs have significant advantages in image processing, as they can automatically learn and extract important features from images. In this step, digestive tract imaging data of patients after endoscopy is first collected. After preprocessing, the data is input into a CNN model. Through multi-layer convolution operations, the CNN can automatically extract lesion features from the images, such as healing bleeding points, ulcer areas, and other potential abnormal changes. By training the CNN model, it can accurately distinguish between normal healing and lesions, thereby identifying abnormal healing lesion features of the digestive tract after endoscopy. During feature recognition, the CNN can also classify according to different lesion degrees to further determine the progress and degree of abnormality in healing. Through the output of this model, the healing lesion features of the digestive tract after endoscopy can be obtained, ultimately yielding the abnormal healing lesion features after endoscopy.

[0105] Furthermore, the simultaneous tracking of the maximum and minimum points corresponding to each physiological indicator curve includes:

[0106] The frequency characteristics of curve fluctuations of each physiological index within the periodic rhythm spectrum of postoperative physiological indicators of digestive endoscopy were analyzed to obtain the frequency characteristics of curve fluctuations of each postoperative physiological index.

[0107] In this embodiment of the invention, time-series data of various physiological indicators, such as heart rate, blood pressure, body temperature, and respiratory rate, are obtained from the post-endoscopic physiological monitoring system. This data needs to be preprocessed, including noise reduction and smoothing, to ensure the accuracy of the original signal. Then, Fourier Transform (FFT) is used to perform frequency domain analysis on these time-series data to calculate the spectrum of each physiological indicator curve. Spectral analysis can reveal the intensity of each frequency component in the signal, thereby identifying the dominant frequency component of periodic fluctuations, i.e., the fluctuation frequency characteristics of each physiological indicator. These frequency characteristics reflect the periodic change law of the physiological system. For example, the fluctuation frequency of heart rate may correspond to the heart's beating rhythm, while body temperature fluctuations are related to the physiological rhythm. Through spectral analysis, the periodic components of each physiological indicator can be extracted, and finally, the curve fluctuation frequency characteristics corresponding to each post-endoscopic physiological indicator are obtained.

[0108] Preferably, based on the curve fluctuation frequency characteristics of each post-endoscopic physiological indicator, the curve fluctuation pattern region of each physiological indicator within the periodic rhythm spectrum of the post-endoscopic physiological indicator is divided to generate a curve fluctuation pattern sub-region corresponding to each post-endoscopic physiological indicator.

[0109] In this embodiment of the invention, the fluctuation frequency characteristics of each physiological indicator are divided into different fluctuation pattern regions based on the previously obtained spectral data. Specifically, a standard for fluctuation pattern division needs to be set first. For example, the spectral data can be classified by statistical analysis methods (such as cluster analysis or threshold-based segmentation) to divide the fluctuations of different frequency bands into different pattern regions. For the spectral graph of each physiological indicator, it can be divided into high-frequency, mid-frequency and low-frequency regions according to the distribution of the dominant frequency component. For example, for heart rate fluctuations, the low-frequency part can correspond to the basic heart rhythm, while the high-frequency part corresponds to the fluctuations reflecting autonomic nerve activity. A similar division is performed on each physiological indicator to obtain the corresponding fluctuation pattern sub-region. This process can simplify and structure the complex fluctuation patterns of each physiological indicator, and finally divide and generate the curve fluctuation pattern sub-regions corresponding to each post-endoscopic physiological indicator.

[0110] Preferably, the golden section search method is used to perform extreme value search processing on the sub-region of the curve fluctuation pattern corresponding to each post-endoscopic physiological indicator, so as to obtain the search interval of the extreme value point of the curve fluctuation corresponding to each post-endoscopic physiological indicator.

[0111] In this embodiment of the invention, the golden section search method is used to locate the extreme points for each previously obtained fluctuation pattern sub-region of physiological indicators. The golden section search method is an efficient optimization method, mainly used to search for the extreme points of a function within a known interval. For each fluctuation pattern sub-region of physiological indicators, the search interval is first determined. This interval is generally determined by the fluctuation pattern region divided in the previous step. Then, the interval is continuously divided into two parts using the golden section method, and the extreme points are determined based on the fluctuation characteristics of the spectrum. By continuously narrowing the search range, the golden section method can accurately locate the vicinity of the maximum and minimum points of the curve. Through this search process, the precise extreme point search interval within the fluctuation pattern sub-region of each physiological indicator can be obtained, and finally, the extreme point search interval of the curve fluctuation corresponding to each post-endoscopic physiological indicator is obtained.

[0112] Preferably, the parabolic approximation method is used to perform curve local extreme point fitting calculation on the search interval of the extreme point of the curve fluctuation corresponding to each postoperative physiological index, so as to obtain the maximum and minimum points corresponding to each physiological index curve.

[0113] In this embodiment of the invention, a parabolic approximation method is used to fit local extreme points of a curve within a previously obtained extreme point search interval. Specifically, the parabolic approximation method is a numerical optimization method mainly used to fit local extreme points on a curve. Using this method, within the extreme point search interval of each physiological indicator's fluctuation pattern sub-region, several points within that interval are first selected, and a quadratic parabola is fitted based on the values ​​of these points. By analyzing the vertex of the parabola, the local maximum or minimum points within that interval can be accurately determined. Since the parabolic approximation method can efficiently calculate the extreme values ​​of a curve, the maximum and minimum points of each physiological indicator curve can be quickly and accurately determined. These extreme points represent the key fluctuation characteristics of the physiological indicator curve within a certain period, reflecting the postoperative physiological state change trend of the patient, and finally obtaining the maximum and minimum points corresponding to each physiological indicator curve.

[0114] Furthermore, the method of using convolutional neural networks to identify postoperative healing lesion features from standard endoscopic gastrointestinal imaging data includes:

[0115] By obtaining the absorption and scattering characteristics of digestive tract tissues under different wavelengths of light through standard images of the digestive tract after endoscopy, and combining the absorption and scattering characteristics of digestive tract tissues under different wavelengths of light with a preset light scattering model, the corresponding standard images of the digestive tract after endoscopy are analyzed for optical parameters to obtain the absorption coefficient and scattering coefficient at each pixel in the images of the digestive tract after endoscopy.

[0116] In this embodiment of the invention, the absorption and scattering coefficients of digestive tract tissues under different wavelengths of light are obtained through standard post-endoscopic digestive tract imaging data. This requires acquiring post-endoscopic images and analyzing them using specific optical parameter models. Specifically, the absorption and scattering characteristics of digestive tract tissues are influenced by tissue type, blood flow, and other physiological characteristics. Based on these tissue optical characteristics, a set of preset light scattering models, such as the Mie scattering model or the Rayleigh scattering model, are selected to simulate the interaction between light and tissue. Then, by combining spectral data and endoscopic images, computer vision technology is used to analyze each pixel of the image, extracting the corresponding absorption and scattering coefficients. The value of each pixel is mapped to specific optical parameters, so that the final image data contains the absorption and scattering characteristics of each pixel at different wavelengths, ultimately obtaining the absorption and scattering coefficients at each pixel in the post-endoscopic digestive tract image.

[0117] Preferably, the image texture fractal structure is identified based on the absorption coefficient and scattering coefficient of each pixel in the post-endoscopic digestive tract image to generate the fractal feature structure of the post-endoscopic digestive tract image, including the fractal feature structure of mucosal folds and vascular branches; the fractal dimension of the post-endoscopic digestive tract image is obtained according to the fractal feature structure of the post-endoscopic digestive tract image.

[0118] In this embodiment of the invention, texture fractal structure recognition of endoscopic image data is performed using absorption and scattering coefficients. First, the optical parameter data of the post-endoscopic image are input into a texture analysis algorithm for fractal feature extraction. Fractal analysis methods, such as box-counting dimension, can be used to identify fractal feature structures in the image. These methods can identify structures such as mucosal folds and vascular branches in the image and calculate their corresponding fractal dimensions. This fractal dimension reflects the complexity and self-similarity of tissue morphology, which helps to quantify the structural features of the digestive tract in subsequent analysis. By analyzing the fractal dimension, information such as the morphological changes of the mucosa and the distribution of the vascular network in the post-endoscopic digestive tract can be further obtained, thereby providing detailed tissue information and finally obtaining the fractal dimension of the post-endoscopic digestive tract image.

[0119] Preferably, by using the gray-level co-occurrence matrix to perform gray-level texture feature analysis on the standard image data of the digestive tract after endoscopy, the gray-level texture feature distribution of the digestive tract image after endoscopy in different directions and between distance pixel pairs is obtained.

[0120] In this embodiment of the invention, gray-level texture feature analysis of post-endoscopic image data is performed using a gray-level co-occurrence matrix (GLCM). Specifically, the standard post-endoscopic image data is converted into a grayscale image, and then the gray-level relationship between each pixel and its surrounding pixels is analyzed. By calculating the gray-level co-occurrence matrix, statistical information on gray-level changes in the image can be obtained, covering features such as texture directionality, contrast, and homogeneity. In this process, by selecting different directions (e.g., 0 degrees, 45 degrees, 90 degrees, and 135 degrees) and different distances (e.g., 1 pixel, 2 pixels, etc.), gray-level texture features under different directions and distances are extracted. The obtained texture features can be used to further analyze the spatial distribution features of the image and provide useful clues to distinguish the different manifestations of healthy tissue and lesion areas. Through the extraction and analysis of these features, better data support can be provided for subsequent anatomical region segmentation and lesion detection, ultimately obtaining the gray-level texture feature distribution of post-endoscopic digestive tract images between pixel pairs at different directions and distances.

[0121] Preferably, based on the fractal dimension of the digestive tract images after endoscopy and the grayscale texture feature distribution of the digestive tract images after endoscopy in different directions and between distance pixel pairs, a semantic segmentation network model is used to segment the corresponding standard image data of the digestive tract after endoscopy to obtain images of the anatomical structures of the digestive tract after endoscopy.

[0122] In this embodiment of the invention, based on the fractal dimension and grayscale texture features obtained in the first two steps, a semantic segmentation network model is used to segment the anatomical regions of the digestive tract images after endoscopy. In this process, the previously extracted absorption coefficient, scattering coefficient, and grayscale texture features are first input into the deep learning semantic segmentation network. The network can achieve accurate identification and segmentation of the digestive tract anatomical regions by combining the spatial and texture features of the image through a convolutional neural network (CNN) structure. Specifically, the network will separate different anatomical regions, such as the stomach wall, mucosa, and blood vessels, according to the texture and fractal features in the image, and assign a label to each region. The segmentation result will yield multiple anatomical structure region images. These regions provide clear boundary and regional information for subsequent lesion detection, healing analysis, etc., and can effectively identify various parts of the digestive tract after surgery, ultimately obtaining images of various anatomical structure regions of the digestive tract after endoscopy.

[0123] Preferably, a convolutional neural network is used to extract postoperative healing lesion features from images of various endoscopic digestive tract anatomical structures, obtaining postoperative bleeding points, ulcer size, and ulcer depth changes. Based on the ulcer depth changes, the corresponding ulcer expansion rate is obtained. Then, based on the ulcer size and expansion rate, the lesion degree is calculated using the ulcer severity formula to measure the corresponding lesion degree of the postoperative digestive tract images, thus obtaining the lesion degree corresponding to the postoperative digestive tract ulcer.

[0124] In this embodiment of the invention, a convolutional neural network (CNN) is used to further analyze the images of various anatomical structures in the digestive tract that were previously segmented after endoscopy, and to extract the features of postoperative healing lesions. The CNN model extracts features from the images of each anatomical region, paying particular attention to changes in the lesion area, such as the identification of healing bleeding points, the measurement of ulcer size, and the calculation of ulcer depth. Through model training, the healing process, lesion areas, and specific manifestations of ulcers in the digestive tract can be accurately identified. Next, based on the changes in the depth of the gastrointestinal ulcer after endoscopy, the expansion rate of the lesion was calculated. For the calculation of the lesion expansion rate, a time series-based analysis method was used to correlate the changes in ulcer area in the endoscopic images with the time axis, thereby calculating the expansion speed of the lesion. By combining the gastrointestinal image area after endoscopy, the size of the ulcer surface of the gastrointestinal lesion after endoscopy, the ulcer area attenuation factor, the ulcer expansion rate of the gastrointestinal lesion after endoscopy, the ulcer expansion rate attenuation factor, the ulcer weighting coefficient after endoscopy, the number of image features, the lesion weighting coefficient of image features, and related parameters, a formula for calculating the degree of ulcer lesion was constructed, and the final lesion metric value was obtained. This lesion metric value can quantify the severity of the gastrointestinal lesion after the operation, and finally obtain the lesion degree corresponding to the gastrointestinal ulcer after endoscopy.

[0125] Furthermore, the formula for calculating the degree of ulcer lesions is as follows:

[0126]

[0127] In the formula, D represents the degree of lesion corresponding to the gastrointestinal ulcer after endoscopy, Ω represents the gastrointestinal imaging area after endoscopy, and A xy R represents the size of the ulcer surface of the digestive tract lesion after endoscopy at location (x, y), λ1 is the ulcer area attenuation factor, and R xy λ1 represents the ulcer expansion rate of the digestive tract lesion at location (x, y) after endoscopy, λ2 is the ulcer expansion rate attenuation factor, α1 is the weighting coefficient of the digestive tract ulcer after endoscopy, n is the number of image features corresponding to the digestive tract image after endoscopy, i is the item index corresponding to the image feature, and f is the ulcer expansion rate at location (x, y).i α1 represents the i-th image feature corresponding to the digestive tract image after endoscopy, α2 is the lesion weighting coefficient of the image feature, and η is the correction coefficient for the degree of lesion.

[0128] This invention, through the use of a specific mathematical model and verification, yields a formula for calculating the severity of ulcer lesions. This formula is used to measure lesions in corresponding post-endoscopic gastrointestinal images. By incorporating multiple influencing factors, such as ulcer area, expansion rate, and image features, this formula comprehensively considers the impact of different lesions, thus obtaining a more accurate lesion severity. Specific influencing factors include: ulcer size, which directly reflects the physical area of ​​the ulcer; generally, a larger ulcer area indicates a more severe lesion; ulcer expansion rate, which measures the growth speed of the ulcer; ulcers with faster growth rates usually represent more severe lesions; and image texture features, which reflect the local structure or changes in the image, reflecting tissue heterogeneity or irregularity and helping to identify ulcer morphological changes. Through the comprehensive analysis of the above variables, various aspects of gastrointestinal ulcers can be better captured, with the ulcer expansion and change process being particularly important. This formula uses weighting coefficients and attenuation factors, which can be adjusted according to the importance of different image regions or features. The weighting coefficients adjust the importance of different features in calculating the overall lesion severity; if the lesion area is more important, the corresponding weighting coefficient can be increased, thus amplifying the area's influence on lesion severity. Similarly, if the image's texture features are more important, the corresponding weighting coefficient can be adjusted. The attenuation factor considers the non-linear effect of ulcer area or expansion rate on lesion severity. As the ulcer area or expansion rate increases, the introduction of the attenuation factor can effectively simulate the complex behavior of ulcers. For example, an increase in area does not necessarily linearly affect lesion severity; a larger area leads to more complex pathological changes. These parameters make the formula more flexible and adaptable to different conditions and image data. Through integral calculations, the formula considers spatial location information; that is, the size and expansion rate of the lesion ulcer are not merely single-point features, but a regional and global information integration. This allows the model to consider overall regional changes rather than local features, thus better simulating real clinical lesions. By incorporating image texture features (such as those extracted from the gray-level co-occurrence matrix) into the calculation of lesion severity, subtle changes in post-endoscopic imaging can be linked to actual histopathological changes. This method helps to detect early or minor lesions, providing more accurate pathological information. Because the formula includes the lesion expansion rate, it can not only assess the current severity of the ulcer but also dynamically track lesion progression, especially during multiple post-endoscopic observations, enabling real-time assessment of whether the lesion is improving or worsening. Furthermore, the correction coefficient in the formula provides further flexibility, allowing for the correction of model biases across different cases or imaging data, ensuring the accuracy of lesion measurement. In summary, this formula fully considers the lesion severity D corresponding to the post-endoscopic gastrointestinal ulcer, the post-endoscopic gastrointestinal imaging region Ω, and the size A of the post-endoscopic gastrointestinal ulcer surface at location (x, y). xyUlcer area attenuation factor λ1, and ulcer expansion rate R of endoscopic gastrointestinal lesions at location (x, y). xy Ulcer expansion rate attenuation factor λ2, post-endoscopic gastrointestinal ulcer weighting coefficient α1, number of image features n corresponding to post-endoscopic gastrointestinal images, item index i corresponding to the image features, and the i-th image feature f corresponding to the post-endoscopic gastrointestinal images. i The image feature lesion weighting coefficient α2 and the lesion severity correction coefficient η, based on the correlation between the lesion severity D corresponding to the gastrointestinal ulcer after endoscopy and the above parameters, constitute a functional relationship. This formula enables the calculation of lesion measurement in corresponding post-endoscopic gastrointestinal images. Furthermore, by introducing a correction coefficient η for the degree of lesion, adjustments can be made based on errors that occur during the calculation process, thereby improving the accuracy and applicability of the formula for calculating the degree of ulcer lesions.

[0129] Furthermore, the post-endoscopic recovery risk assessment module includes the following functions:

[0130] The mean and variance of the physiological indicators and abnormal lesions after endoscopic surgery were obtained by measuring the long and short time-series characteristics of physiological indicators and the abnormal lesions after endoscopic surgery. Based on the mean and variance of the characteristics, Z-score standardization was performed on the physiological indicators and abnormal lesions after endoscopic surgery to obtain the standard characteristics of physiological indicators and abnormal lesions after endoscopic surgery.

[0131] In this embodiment of the invention, data preprocessing is performed on the long- and short-term temporal characteristics of physiological indicators and the characteristics of abnormal healing lesions in patients after endoscopic surgery. The long- and short-term temporal characteristics of physiological indicators refer to the data on changes in heart rate, respiratory rate, body temperature, blood pressure, etc., of patients after surgery over time. The characteristics of abnormal healing lesions include imaging data on the healing progress of the gastrointestinal tract, bleeding, scar tissue, etc., of patients after endoscopic surgery. For these characteristics, the mean and variance of each feature are first calculated. These two statistics are used to reflect the central tendency and dispersion of the data. Then, for each feature, the Z-score standardization method is used to standardize the data. Specifically, for each data point, the Z-score standardization formula is: Standardized feature value = (Original data point - Feature mean) / Feature standard deviation. Finally, the long- and short-term standard characteristics of physiological indicators and the standard characteristics of abnormal lesions after endoscopic surgery are obtained.

[0132] Preferably, mutual information analysis is performed on the various dimensions of the standard features of physiological indicators after endoscopy and the various feature components of the standard features of abnormal lesions after endoscopy to obtain the mutual information between the various physiological dimensions and the various abnormal lesion feature components after endoscopy; based on the mutual information between the various physiological dimensions and the various abnormal lesion feature components after endoscopy, nonlinear correlation evaluation analysis is performed on the various dimensions of the standard features of physiological indicators after endoscopy and the various feature components of abnormal lesions after endoscopy to obtain the nonlinear correlation between the various physiological dimensions and the various abnormal lesion feature components after endoscopy.

[0133] In this embodiment of the invention, mutual information analysis is performed on the previously standardized physiological indicators (length and short standard features) and the abnormal lesion healing standard features. Mutual information is a measure of the correlation between two variables and can assess the interdependence between different dimensional features and lesion features. In the specific operation, a suitable algorithm (such as an entropy-based mutual information calculation method) is first selected to calculate the mutual information value between each pair of physiological features and abnormal lesion features. The calculation process can use the sklearn.metrics.mutual_info_score function in Python or a custom mutual information calculation formula. After calculating the mutual information, the mutual information value between each pair of physiological indicators and abnormal lesion features is obtained. These values ​​reflect the dependency relationship between different features. Based on these mutual information values, nonlinear correlation analysis is further performed to assess the complex nonlinear dependency relationship between physiological features and lesion features, and finally, the nonlinear correlation between each physiological dimension feature and each abnormal lesion feature component after endoscopic surgery is obtained.

[0134] Preferably, a linear correlation assessment analysis is performed on the various dimensions of the physiological indicators after endoscopy and the various components of the abnormal lesion standard characteristics after endoscopy to obtain the linear correlation coefficient between each physiological dimension and each abnormal lesion component after endoscopy.

[0135] In this embodiment of the invention, the linear correlation between postoperative physiological indicators and abnormal lesion characteristics after endoscopy is further analyzed in depth, and calculated using the Pearson Correlation Coefficient. This method is widely used to assess the linear relationship between two variables. Specifically, for each pair of physiological dimension features and abnormal lesion feature components, the Pearson Correlation Coefficient is calculated based on the previously obtained standardized feature data. The calculation formula is as follows: Where X u and Y u These represent the values ​​for physiological characteristics and abnormal pathological characteristics, respectively. and The mean of these features is used to obtain the linear correlation coefficient, which reflects the strength of the linear relationship between physiological and pathological features. Based on the magnitude of the correlation coefficient, the impact of different features on postoperative recovery can be further evaluated, and finally, the linear correlation coefficient between each physiological dimension feature and each abnormal pathological feature component after endoscopic surgery can be obtained.

[0136] Preferably, the long and short time-series features of physiological indicators after endoscopy and the features of abnormal lesions after endoscopy are input into a preset deep neural network model, and the nonlinear correlation and linear correlation coefficient between the features of each physiological dimension after endoscopy and the features of each abnormal lesion are combined to perform feature fusion to generate a fused feature set after digestive endoscopy.

[0137] In this embodiment of the invention, a deep learning model is used to fuse the long and short temporal features of physiological indicators and the features of abnormal healing lesions. First, a suitable deep neural network architecture (such as a multilayer perceptron, convolutional neural network, etc.) is selected for this task, and the previously obtained standardized features are input into the model. The design of the deep neural network takes into account the nonlinear correlation between features. Through multilayer nonlinear transformation, the model can automatically capture the complex relationship between different features. When inputting data, the input features are weighted and fused by combining the previously obtained mutual information value and the calculated linear correlation coefficient. The goal of this weighted fusion is to assign different weights to different features according to the correlation between each feature, so that the model can more effectively learn the feature information that has an important impact on the prediction of postoperative recovery. In this way, the generated fused feature set can comprehensively reflect the potential risk factors of the patient's postoperative recovery, and finally generate the postoperative fused feature set of digestive endoscopy.

[0138] Preferably, the postoperative recovery risk assessment calculation formula is used to calculate the postoperative recovery risk of the fusion feature set after digestive endoscopy, so as to obtain the degree of recovery risk after digestive endoscopy.

[0139] In this embodiment of the invention, a suitable postoperative recovery risk assessment calculation formula is constructed by combining the postoperative observation time range, time variable parameters, postoperative physiological index time series characteristics, healing abnormal lesion time series characteristics, healing abnormal lesion recovery risk weight coefficient, postoperative physiological index long and short time series characteristics, postoperative physiological index recovery risk weight coefficient, postoperative recovery risk influencing factors of digestive endoscopy, specific values ​​corresponding to the individual characteristics of patients after digestive endoscopy, weight coefficients and related parameters. This formula is used to calculate the postoperative recovery risk of the fusion feature set after digestive endoscopy, so as to quantify and predict the degree of postoperative recovery risk of patients, and finally obtain the degree of postoperative recovery risk of digestive endoscopy.

[0140] Furthermore, the specific formula for calculating the postoperative recovery risk assessment is as follows:

[0141]

[0142] In the formula, R represents the risk level of recovery after digestive endoscopy, T represents the postoperative observation time range, t represents the time variable parameter, P(t) represents the time series characteristics of postoperative physiological indicators at time point t, Q(t) represents the time series characteristics of abnormal healing lesions at time point t, β represents the risk weighting coefficient for recovery of abnormal healing lesions, H(t) represents the long-term time series characteristics of postoperative physiological indicators at time point t, γ represents the risk weighting coefficient for recovery of postoperative physiological indicators, ε represents the risk influencing factor for recovery after digestive endoscopy, m represents the total number of individual patient characteristics after digestive endoscopy, and X represents the risk factor. j φ is the specific numerical value corresponding to the individual characteristics of the j-th patient after digestive endoscopy. j ξ represents the weighting coefficient corresponding to the individual characteristics of the j-th patient after digestive endoscopy, and ξ represents the correction coefficient for the degree of recovery risk after digestive endoscopy.

[0143] This invention, through the use of a specific mathematical model and verification, derives a postoperative recovery risk assessment formula for calculating the postoperative recovery risk of a fusion feature set after digestive endoscopy. In this formula, P(t) and Q(t) represent the characteristics of the patient's postoperative physiological indicators and abnormal healing lesions in a time series. Using this time series data, dynamic changes during the postoperative recovery process can be captured. By performing time series analysis on the characteristics at different postoperative time points, the patient's recovery trend can be dynamically assessed, and postoperative risk changes can be identified in real time, ensuring early intervention. Individual characteristics, along with corresponding weighting coefficients, which can include the patient's clinical background, health status, etc., work together with other physiological and pathological characteristics to influence postoperative recovery. By weighting and integrating individual characteristics, the formula provides a quantitative, personalized risk assessment. This allows for a more precise tailoring of recovery expectations for each patient, ensuring that individual differences are fully considered. In the formula, β represents the recovery risk weighting coefficient for abnormal healing lesion characteristics, indicating the degree of influence of the risk of abnormal healing lesions on postoperative recovery. Different lesion types or recovery processes have different weight values. This weighting mechanism allows for more precise attention to abnormal lesions in different behaviors during the healing process (such as infection, bleeding, scarring, etc.), ensuring effective monitoring and adjustment of high-risk factors in postoperative recovery. Furthermore, the correction coefficient in the formula provides a final calibration of the recovery risk level. This adjustment is made for certain special circumstances or external influencing factors (such as patient comorbidities, emergencies, etc.). The introduction of the correction coefficient allows for flexible adjustment of risk assessment results based on external factors or unforeseen changes, making the assessment more accurate and operable. In summary, this formula fully considers the risk level R of postoperative recovery after digestive endoscopy, the postoperative observation time range T, the time variable parameter t, the time series characteristics P(t) of the postoperative physiological indicators at time point t, the time series characteristics Q(t) of the abnormal healing lesions at time point t, the risk weighting coefficient β of the abnormal healing lesions, the long and short time series characteristics H(t) of the postoperative physiological indicators at time point t, the risk weighting coefficient γ of the postoperative physiological indicators, the risk influencing factor ε of postoperative recovery after digestive endoscopy, the total number m of individual patient characteristics after digestive endoscopy, and the specific value X of the individual patient characteristic after the j-th digestive endoscopy. j The weighting coefficient φ corresponding to the individual characteristics of the j-th patient after digestive endoscopy. j The correction coefficient ξ for the recovery risk level after digestive endoscopy is based on a functional relationship between the recovery risk level R after digestive endoscopy and the above parameters:

[0144]

[0145] This formula enables the calculation of postoperative recovery risk assessment based on the fusion feature set after digestive endoscopy. Furthermore, by introducing a correction coefficient ξ for the degree of recovery risk after digestive endoscopy, adjustments can be made based on errors that occur during the calculation process, thereby improving the accuracy and applicability of the postoperative recovery risk assessment formula.

[0146] Furthermore, the postoperative recovery risk intelligent care module includes the following functions:

[0147] Acquire clinical experience after digestive endoscopy and determine a risk grading system for postoperative recovery based on this experience.

[0148] In this embodiment of the invention, by analyzing a large amount of clinical data from patients after digestive endoscopy, multi-dimensional data such as surgery type, patient basic information (e.g., age, gender, medical history), and postoperative recovery status are collected. This data can come from the hospital's electronic health record (EHR) system or clinical database. After data cleaning and preprocessing, key factors related to postoperative recovery are extracted, such as indicators of bleeding, infection, pain control, and gastrointestinal function recovery. These data are trained using machine learning algorithms, especially deep learning models (e.g., convolutional neural networks (CNN) or recurrent neural networks (RNN). The model can identify the main factors affecting postoperative recovery and construct a predictive model of recovery risk based on these factors. In this process, the model models the relationships between various variables and provides different recovery risk classification standards based on the performance in historical cases. The generated recovery risk classification system can subdivide the patient's recovery process into multiple stages and classify patients according to their recovery status, for example, classifying them into three levels: low risk, medium risk, and high risk, thus ultimately determining the postoperative recovery risk classification system after endoscopy.

[0149] Preferably, the recovery risk level of patients after digestive endoscopy is determined based on the recovery risk grading system after endoscopic surgery, including low recovery risk level, medium recovery risk level and high recovery risk level.

[0150] In this embodiment of the invention, the actual situation of each postoperative patient after digestive endoscopy is compared with the standards in the previously established recovery risk grading system. Specifically, a trained deep learning model or other machine learning model is used, inputting the patient's basic postoperative information and clinical data. Based on the features extracted by the model and the established risk grading standards, the patient's recovery risk is automatically assessed. For example, the model can analyze factors such as postoperative complications, postoperative temperature changes, pain scores, and bowel movements to determine whether the patient belongs to a low, medium, or high recovery risk level. If the patient recovers well post-surgery with no significant complications and their bowel movements and appetite gradually return to normal (i.e., the risk level is 0%-45%), they are assessed as low-risk. If the patient experiences mild complications, recovers slowly, or requires medication (i.e., the risk level is 46%-85%), they are assessed as medium-risk. If the patient experiences severe complications such as severe bleeding, infection, or functional impairment (i.e., the risk level is 86%-100%), they are assessed as high-risk. This step is automated by an algorithm to determine the recovery risk level, ultimately yielding the corresponding recovery risk level for the patient after digestive endoscopy.

[0151] Preferably, postoperative intelligent nursing analysis is performed based on the recovery risk level of the patient after digestive endoscopy to generate recovery nursing suggestions for the patient after digestive endoscopy, so as to carry out the corresponding postoperative risk recovery nursing work.

[0152] In this embodiment of the invention, postoperative care for each patient is intelligently analyzed based on a previously determined recovery risk level. Specifically, according to the patient's recovery risk level (low, medium, high), a pre-established nursing intervention model is used to generate personalized nursing recommendations for the patient. For example, for low-risk patients, nursing recommendations may include routine monitoring, appropriate activity guidance, and dietary advice; for medium-risk patients, nursing recommendations involve more detailed monitoring, medication, and timely interventions; for high-risk patients, nursing recommendations need to include close monitoring of vital signs, early detection of complications and preventative measures, and timely postoperative follow-up examinations, etc., and will be tailored to the patient's individual circumstances. Based on specific circumstances (such as postoperative bleeding risk, infection risk, and gastrointestinal function recovery status), a specific nursing plan is generated using clinical experience rules. This process leverages the reasoning capabilities of deep learning models, combined with the clinical experience rules of medical experts, to provide customized and precise nursing plans in a data-driven manner. This intelligent nursing system can also make real-time adjustments based on the dynamic changes during the patient's recovery process. For example, if the patient's recovery status changes, it can provide real-time feedback and adjust the nursing plan to ensure that nursing measures always match the patient's recovery status. Ultimately, it generates postoperative recovery nursing recommendations for patients after digestive endoscopy to execute corresponding postoperative risk recovery nursing procedures.

[0153] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features of the invention herein.

Claims

1. A deep learning-based intelligent nursing system for postoperative care after digestive endoscopy, characterized in that, Includes the following modules: The post-endoscopic data acquisition and processing module is used to acquire post-endoscopic physiological data and post-endoscopic digestive tract imaging data of patients through the API interface, and to perform post-endoscopic data preprocessing and standardization on the post-endoscopic physiological data and post-endoscopic digestive tract imaging data to obtain post-endoscopic physiological standardized data and post-endoscopic digestive tract standardized imaging data. The post-endoscopic feature analysis module is used to extract long and short-term features from standardized physiological data after digestive endoscopy using a long short-term memory network to obtain long and short-term features of post-endoscopic physiological indicators; and to identify post-operative healing lesion features from standardized digestive tract imaging data after endoscopy using a convolutional neural network to obtain abnormal healing lesion features after endoscopy, including the degree of lesion corresponding to post-endoscopic digestive tract bleeding, ulceration, and infection. The post-endoscopic feature analysis module includes the following functions: After the physiological standardization data of digestive endoscopy, the frequency domain transformation of physiological indicators was performed to obtain the periodic rhythm spectrum of physiological indicators after digestive endoscopy. A dynamic trend key node analysis of the circadian rhythm spectrum of post-endoscopic physiological indicators was performed using a slope change detection and extreme point tracking method. This involved analyzing the slope changes of each physiological indicator curve along the time axis. When a sudden slope change exceeded a preset threshold, it was marked as a potential key node. Simultaneously, the maximum and minimum values ​​of each physiological indicator curve were tracked to obtain a set of dynamic trend key nodes for post-endoscopic physiological indicators. The simultaneous tracking of the maximum and minimum values ​​of each physiological indicator curve included: The frequency characteristics of curve fluctuations of each physiological index within the periodic rhythm spectrum of postoperative physiological indicators of digestive endoscopy were analyzed to obtain the frequency characteristics of curve fluctuations of each postoperative physiological index. Based on the curve fluctuation frequency characteristics of each postoperative physiological indicator, the curve fluctuation pattern region of each physiological indicator within the periodic rhythm spectrum of postoperative physiological indicators is divided to generate the curve fluctuation pattern sub-region corresponding to each postoperative physiological indicator. The golden section search method was used to perform extreme value search on the sub-region of the curve fluctuation pattern corresponding to each post-endoscopic physiological index, and the search interval of the extreme value point of the curve fluctuation corresponding to each post-endoscopic physiological index was obtained. The parabolic approximation method was used to perform curve local extreme point fitting calculation on the search interval of the extreme point of the curve fluctuation corresponding to each postoperative physiological index, so as to obtain the maximum and minimum points corresponding to each physiological index curve. Based on the dynamic trend of post-endoscopic physiological indicators, long-scale time windows and short-scale time windows are determined on the time axis within the periodic rhythm spectrum of post-endoscopic physiological indicators between each two adjacent key nodes in the key node set. Based on the long-scale time windows and short-scale time windows, the corresponding post-endoscopic physiological standardized data are divided into long-scale and short-scale window sequences to obtain the long-scale physiological sequence data and the short-scale physiological sequence data of endoscopic surgery. Long Short-Term Memory (LSTM) networks were used to extract long and short time-series features from long-term and short-term physiological window sequence data of endoscopic surgery to obtain long and short time-series features of postoperative physiological indicators, including low-frequency trend change features of each physiological indicator under the long time window and high-frequency fluctuation features under the short time window. Convolutional neural networks were used to identify postoperative healing lesion features from standard endoscopic gastrointestinal imaging data, resulting in abnormal postoperative healing lesion features, including postoperative gastrointestinal bleeding points and the degree of lesions corresponding to ulcers. The postoperative recovery risk assessment module is used to input the long- and short-term temporal characteristics of postoperative physiological indicators and the characteristics of abnormal lesions after postoperative healing into a preset deep neural network model for feature fusion to generate a fused feature set after digestive endoscopy; the postoperative recovery risk assessment is calculated on the fused feature set after digestive endoscopy to obtain the degree of recovery risk after digestive endoscopy. The intelligent nursing module for postoperative recovery risk is used to determine the recovery risk level of patients after digestive endoscopy based on the degree of recovery risk after digestive endoscopy, and to perform intelligent postoperative nursing analysis based on the recovery risk level of patients after digestive endoscopy to generate recovery nursing suggestions for patients after digestive endoscopy, so as to carry out corresponding postoperative risk recovery nursing operations.

2. The intelligent postoperative nursing system for digestive endoscopy based on deep learning according to claim 1, characterized in that, The post-endoscopic data acquisition and processing module includes the following functions: By using blockchain-based encryption technology and digital certificate authentication, a secure data collection channel is built between the hospital's internal network and external data requesters. This channel records the corresponding postoperative data interaction logs of patients after digestive endoscopy through the distributed ledger of the blockchain and uses digital certificates to perform identity authentication on external data requesters, thus generating a secure data collection channel for postoperative digestive endoscopy. The characteristics of postoperative patient data from digestive endoscopy are obtained, including the storage precision and timestamp format of the patient's physiological data, as well as the resolution and encoding method of the endoscopic image data. The data structure of the hospital information system and the digestive endoscopy image storage system is combined to finely adapt the API interface parameters of the external data request end to generate a set of data request adapted API interface parameters, including data filtering conditions and data format conversion rules. Based on the secure data acquisition channel after digestive endoscopy and the API interface parameter set adapted to the data request, the authorized external data request terminal sends acquisition requests to the corresponding hospital information system and digestive endoscopy image storage system in the hospital's internal network to obtain the patient's physiological data and digestive tract status image data after digestive endoscopy. Postoperative data preprocessing and standardization were performed on physiological data and gastrointestinal imaging data of patients after digestive endoscopy to obtain standardized physiological data and standardized gastrointestinal imaging data after digestive endoscopy.

3. The intelligent postoperative nursing system for digestive endoscopy based on deep learning according to claim 2, characterized in that, The postoperative data preprocessing and standardization of physiological data and postoperative digestive tract imaging data of patients after digestive endoscopy includes: Abnormalities were removed and standardized from the physiological data of patients after digestive endoscopy to obtain standardized physiological data after digestive endoscopy. Image data enhancement was performed on the digestive tract status images after endoscopy, including rotation, translation and flipping, to obtain enhanced digestive tract images after endoscopy. Gray-level histogram equalization was performed on the enhanced gastrointestinal images after endoscopy to obtain contrast-balanced images of the gastrointestinal tract after endoscopy. Pixel blurring analysis was performed on the contrast-balanced images of the digestive tract after endoscopy to obtain the pixel blurring values ​​of the digestive tract images after endoscopy; based on the pixel blurring values ​​of the digestive tract images after endoscopy, the corresponding contrast-balanced images of the digestive tract after endoscopy were denoised to obtain denoised images of the digestive tract after endoscopy. The denoised digestive tract images after endoscopy were standardized to obtain standard digestive tract images after endoscopy.

4. The intelligent postoperative nursing system for digestive endoscopy based on deep learning according to claim 1, characterized in that, The method of using convolutional neural networks to identify postoperative healing lesion features from standard endoscopic gastrointestinal imaging data includes: By obtaining the absorption and scattering characteristics of digestive tract tissues under different wavelengths of light through standard images of the digestive tract after endoscopy, and combining the absorption and scattering characteristics of digestive tract tissues under different wavelengths of light with a preset light scattering model, the corresponding standard images of the digestive tract after endoscopy are analyzed for optical parameters to obtain the absorption coefficient and scattering coefficient at each pixel in the images of the digestive tract after endoscopy. Based on the absorption and scattering coefficients of each pixel in the post-endoscopic digestive tract image, the corresponding post-endoscopic digestive tract standard image data are used to identify the image texture fractal structure to generate the fractal feature structure of the post-endoscopic digestive tract image, including the fractal feature structure of mucosal folds and vascular branches; the fractal dimension of the post-endoscopic digestive tract image is obtained according to the fractal feature structure of the post-endoscopic digestive tract image. By using the gray-level co-occurrence matrix to perform gray-level texture feature analysis on standard endoscopic digestive tract images, the distribution of gray-level texture features of endoscopic digestive tract images in different directions and between pixel pairs at different distances was obtained. Based on the fractal dimension of post-endoscopic digestive tract images and the gray-scale texture feature distribution of post-endoscopic digestive tract images in different directions and between distance pixel pairs, a semantic segmentation network model is used to segment the corresponding post-endoscopic digestive tract standard image data into image anatomical regions, thereby obtaining images of various post-endoscopic digestive tract anatomical structure regions. Convolutional neural networks were used to extract postoperative healing lesion features from images of various endoscopic gastrointestinal anatomical structures, obtaining postoperative bleeding points, ulcer size, and ulcer depth changes. Based on the ulcer depth changes, the corresponding ulcer expansion rate was obtained. Then, based on the ulcer size and expansion rate, the lesion severity was calculated using a formula to measure the lesion severity of the corresponding postoperative gastrointestinal images, yielding the severity of the ulcers.

5. The intelligent postoperative nursing system for digestive endoscopy based on deep learning according to claim 4, characterized in that, The specific formula for calculating the severity of the ulcer lesion is as follows: ； In the formula, This refers to the degree of lesion corresponding to gastrointestinal ulcers after endoscopy. This is the area of ​​the digestive tract as shown in the post-endoscopic imaging. For in position Size of ulcers in the digestive tract following endoscopic surgery. As a decay factor for ulcer area size, For in position The rate of ulceration expansion in the digestive tract after endoscopy. As a factor that attenuates the rate of ulcer expansion, This represents the weighted coefficient for post-endoscopic gastrointestinal ulcers. This represents the number of image features corresponding to the digestive tract images after endoscopy. This is the item index corresponding to the image feature. The first image corresponding to the digestive tract after endoscopy Image features, The weighting coefficients for image feature lesions. This is a correction factor for the severity of the lesion.

6. The intelligent postoperative nursing system for digestive endoscopy based on deep learning according to claim 1, characterized in that, The post-endoscopic recovery risk assessment module includes the following functions: The mean and variance of the physiological indicators and abnormal lesions after endoscopic surgery were obtained by measuring the long and short time-series characteristics of physiological indicators and the abnormal lesions after endoscopic surgery. Based on the mean and variance of the characteristics, Z-score standardization was performed on the physiological indicators and abnormal lesions after endoscopic surgery to obtain the standard characteristics of physiological indicators and abnormal lesions after endoscopic surgery. Mutual information analysis was performed on the various dimensions of the standard features of physiological indicators after endoscopy and the various feature components of the standard features of abnormal lesions after endoscopy to obtain the mutual information between the various physiological dimensions and the various abnormal lesion feature components after endoscopy. Based on the mutual information between the various physiological dimensions and the various abnormal lesion feature components after endoscopy, nonlinear correlation assessment analysis was performed on the various dimensions of the standard features of physiological indicators after endoscopy and the various feature components of abnormal lesions after endoscopy to obtain the nonlinear correlation between the various physiological dimensions and the various abnormal lesion feature components after endoscopy. A linear correlation assessment analysis was performed on the various dimensions of the standard features of physiological indicators after endoscopy and the various feature components of the standard features of abnormal lesions after endoscopy, and the linear correlation coefficients between the various physiological dimensions and the feature components of abnormal lesions after endoscopy were obtained. The long and short time-series features of physiological indicators after endoscopy and the features of abnormal lesions after endoscopy are input into a preset deep neural network model. The nonlinear correlation and linear correlation coefficient between the physiological features of each dimension after endoscopy and the features of each abnormal lesion are combined to perform feature fusion to generate a fused feature set after digestive endoscopy. The postoperative recovery risk assessment formula was used to calculate the postoperative recovery risk of the fusion feature set after digestive endoscopy, and the degree of recovery risk after digestive endoscopy was obtained.

7. The intelligent postoperative nursing system for digestive endoscopy based on deep learning according to claim 6, characterized in that, The specific formula for calculating the postoperative recovery risk assessment is as follows: ； In the formula, To assess the risk level of recovery after digestive endoscopy, This refers to the postoperative observation period. For time-varying parameters, For patients at postoperative time points The time series characteristics of postoperative physiological indicators at the corresponding time points. For patients at postoperative time points The time series characteristics of the corresponding abnormal healing lesions Risk weighting coefficient for recovery of abnormal lesions. For patients at postoperative time points The time-series characteristics of postoperative physiological indicators corresponding to the time period. The risk weighting coefficient for the recovery of postoperative physiological indicators. Risk factors influencing recovery after digestive endoscopy This represents the total number of patients with individual characteristics following digestive endoscopy. For the first Specific numerical values ​​corresponding to the individual characteristics of each patient after digestive endoscopy. For the first Weighting coefficients corresponding to individual characteristics of patients after digestive endoscopy. This is a correction factor for the risk of recovery after digestive endoscopy.

8. The intelligent postoperative nursing system for digestive endoscopy based on deep learning according to claim 1, characterized in that, The intelligent nursing module for postoperative recovery risks includes the following functions: Acquire clinical experience after digestive endoscopy and determine a risk grading system for postoperative recovery based on this experience. Based on the post-endoscopic recovery risk grading system, the recovery risk level of patients after digestive endoscopy is determined according to the degree of recovery risk after digestive endoscopy, including low recovery risk level, medium recovery risk level and high recovery risk level. Postoperative intelligent nursing analysis is performed based on the recovery risk level of patients after digestive endoscopy to generate recovery nursing suggestions for patients after digestive endoscopy, so as to carry out corresponding risk recovery nursing tasks after digestive endoscopy.

Images