Remote nursing method and system for patients in digestive system department

Abstract

The invention discloses a remote nursing method and system for patients in the digestive system department, and relates to the technical field of medical nursing. Based on mucous membrane healing quantitative evaluation of artificial intelligence and three-dimensional space registration, subjective experience is converted into objective data flow, and through a deep learning segmentation model, an objective data flow is obtained; a focus area in an endoscope image is automatically identified and segmented, endoscope sequences of previous examinations of a patient are reconstructed and registered to the same three-dimensional intestinal coordinate system, and on the unified space reference, the change of a first quantitative index, a second quantitative index and a third quantitative index along with time is dynamically presented in the form of a curve graph, so that quantitative tracking of remote monitoring precision is realized; according to the quantitative data based on the specific focus identity, adjustment of the treatment strategy can be established on the basis of tiny and measurable changes, the diagnosis and treatment accuracy is greatly improved, and doctors can visually judge the healing rate and fluctuation and recognize the risk of poor treatment response in advance.

Description

Technical Field

[0001] This invention relates to the field of medical and nursing technology, specifically to a remote nursing method and system for patients in the gastroenterology department. Background Technology

[0002] Inflammatory bowel disease (IBD) is a chronic, lifelong inflammatory disease of the digestive tract. Its diagnosis, treatment assessment, and long-term management heavily rely on endoscopy. Mucosal healing has become a recognized key goal in IBD treatment, serving as a powerful indicator for predicting long-term clinical remission, reducing hospitalization rates, and minimizing surgical risks. However, traditional assessments of mucosal healing rely entirely on the visual observation and subjective judgment of gastroenterologists during colonoscopy, employing semi-quantitative scales such as the Mayo Cognitive Assessment (MCA) score or the Crohn's Disease Simplified Endoscopic Assessment (SIA). While these scales standardize the assessment to some extent, they are essentially coarse-grained classifications based on grades, failing to accurately reflect the subtle, continuous changes in mucosal condition. Furthermore, they are heavily influenced by physician experience and subjectivity, leading to significant inter-observer variability in assessment results.

[0003] In existing technologies, AI-assisted endoscopic diagnosis focuses on solving the qualitative identification problem of a single examination, and cannot meet the need for quantitative tracking of the dynamic evolution of chronic diseases. For example, CN110974179A discloses a deep learning-based auxiliary diagnostic system for early gastric cancer, which uses an improved YOLOv3 network to identify and label early cancer lesions in single-frame electronic chromoendoscopic images in real time, solving the problems of accuracy and real-time performance in immediate lesion detection. Meanwhile, CN111524124A discloses an AI-assisted system for digestive endoscopy images in inflammatory bowel disease, which uses a MedicalCnn neural network model to perform binary classification on single colonoscopy images, achieving auxiliary identification of disease types. However, its functions are limited to the immediate identification and qualitative classification of lesions in a single examination, and it lacks the ability to track and quantify the longitudinal temporal evolution of the same specific lesion over multiple examinations spanning several weeks. Secondly, its analysis dimension is limited to qualitative or semi-quantitative scoring of two-dimensional images, and it cannot extract key morphological parameters of lesions in three-dimensional space. It also fails to integrate with the patient's long-term data stream, which leads to doctors still facing the dilemma of fragmented information and subjective and difficult historical comparison when evaluating efficacy and guiding remote management.

[0004] Therefore, both clinical practice and existing technologies indicate an urgent need for a remote nursing system that can overcome the limitations of a single examination and enable long-term, multi-dimensional quantitative tracking of the mucosal healing process, thereby transforming subjective and vague follow-up assessments into objective and precise data-driven decision-making. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a remote nursing method and system for gastroenterology patients. It can quantitatively assess mucosal healing based on artificial intelligence and three-dimensional spatial registration, transforming subjective experience into objective data streams. Through a deep learning segmentation model, it automatically identifies and segments lesion areas in endoscopic images, reconstructs and registers the patient's endoscopic sequences from previous examinations to the same three-dimensional intestinal coordinate system. On this unified spatial benchmark, the changes of the first, second, and third quantitative indicators over time are dynamically presented as curves, achieving quantitative tracking of remote monitoring accuracy. This quantitative data based on specific lesion identities allows adjustments to treatment strategies to be based on minute, measurable changes, greatly improving the accuracy of diagnosis and treatment. Doctors can intuitively judge the healing rate and fluctuations, and identify the risk of poor treatment response in advance. This intuitive cross-period comparison based on a unified coordinate system, combined with trend analysis, enables the development of more targeted and individualized subsequent treatment and follow-up plans.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: On one hand, a remote nursing method for patients in the gastroenterology department, the specific steps of which are as follows:

[0007] S100. Construct and maintain an endoscopy image time-series database to store video and image data of a patient’s past endoscopic examinations. Each data is associated with the corresponding patient identity information and examination timestamp.

[0008] S200: Call the deep learning segmentation model to process the endoscopic images in the endoscopic image time series database, automatically segment the lesion area in the image, and register the images at different time points in three-dimensional space to determine the spatial correspondence of the same lesion at different time points.

[0009] S300. For each lesion that has been registered in S200, calculate the quantitative indicator set for each lesion. The quantitative indicator set includes a first quantitative indicator, a second quantitative indicator, and a third quantitative indicator. Associate the quantitative indicator set with the examination timestamp to form a time-series quantitative data sequence of the lesion.

[0010] S400: Based on the generated time-series quantitative data sequence, graphically display the change curves of the first quantitative index, the second quantitative index and the third quantitative index of each lesion over time, and link to display the lesion image comparison view at different time points.

[0011] S500: The doctor, in conjunction with the change curve and the lesion image comparison view, formulates / adjusts the nursing plan and sends it to the patient. The patient provides feedback on the implementation status, and the information is synchronized to the medical staff.

[0012] Furthermore, in S200, the deep learning segmentation model is a U-Net network based on an encoder-decoder architecture, which comprises:

[0013] The encoder path consists of multiple convolutional layers and max pooling layers, used to extract feature maps layer by layer from the input endoscopic images and compress the features to capture the contextual semantic information of the lesions.

[0014] Decoder path: Consists of multiple deconvolutional layers and upsampling layers, used to restore the feature maps extracted by the encoder to the spatial resolution of the original input image for pixel-level localization;

[0015] Final output layer: The features output by the decoder path are mapped to a single-channel probability map of the same size as the input image. The value of each pixel in the single-channel probability map represents the probability that it belongs to the lesion region.

[0016] Furthermore, in step S200, the specific steps for automatically segmenting the lesion region in the image are as follows:

[0017] The endoscopic images to be processed are input into the deep learning segmentation model;

[0018] The encoder path of the deep learning segmentation model extracts and downsamples the input endoscopic image to obtain image features containing high-level semantic information.

[0019] The decoder path upsamples and refines image features, and restores spatial accuracy by fusing features of the same scale from the encoder path.

[0020] The final output layer generates a probability map. By setting a probability threshold, pixels with probability values ​​higher than the probability threshold in the probability map are identified as lesion pixels, thereby generating a binary lesion mask image and completing the automatic segmentation of various types of lesions such as ulcers, erosions, and edema.

[0021] Furthermore, in step S200, images from different time points are registered in three-dimensional space. The registration process is as follows:

[0022] The endoscopic video sequence was reconstructed to reconstruct a sparse three-dimensional point cloud of the intestine and the camera pose during the examination.

[0023] The 3D point cloud reconstructed in the current inspection is registered with the historical 3D point cloud used as a reference through the iterative nearest point algorithm to find the spatial transformation matrix that minimizes the average distance between the two point clouds.

[0024] Using the spatial transformation matrix, all currently examined image data and their corresponding lesion segmentation results are uniformly mapped to the historical three-dimensional coordinate system of the reference, thereby achieving precise spatial alignment of images at different time points.

[0025] Furthermore, in S300, the first quantification index is the area of ​​the lesion. On the completed three-dimensional registration, the set of three-dimensional mesh vertices covered by the lesion mask image is obtained. By calculating the sum of the surface areas of the three-dimensional mesh, the actual projected area of ​​the lesion in three-dimensional space is obtained.

[0026] Furthermore, in step S300, the second quantification indicator is the depth of the lesion. Based on the completed 3D registration, the 3D point cloud data of the lesion region is extracted. Extracting three-dimensional point cloud data of normal mucosal areas within a preset range around the lesion. ,pass The depth of the lesion was calculated. Among them, Avg Point cloud representing normal mucosal area The average of the elevation values ​​of all points Point cloud representing the lesion area The average elevation value of all points.

[0027] Furthermore, in S300, the third quantification index is the vascular texture score of the lesion. Based on the completed 3D registration, using the segmented lesion boundary as a benchmark, a pre-defined annular region is expanded as the mucosal region of interest (ROI) for vascular texture analysis. This ensures the analysis area is adjacent to the lesion and excludes interference from the lesion's own irregular structure. The ROI undergoes contrast-limited adaptive histogram equalization to enhance the contrast between the vascular structure and the surrounding mucosa and reduce the impact of uneven illumination. Based on the histogram-equalized image, the gray-level co-occurrence matrix of the ROI is calculated. This gray-level co-occurrence matrix describes the image's gray levels. The pixels and grayscale are The probability of pixels co-occurring is used to extract texture features from the gray-level co-occurrence matrix, including correlation, entropy, and second moment of angle. The average value of the texture features is extracted as a comprehensive feature quantity, and the comprehensive feature quantity is synthesized into a single first vascular texture index through linear weighting.

[0028] On the other hand, a remote care system for patients in the gastroenterology department includes a patient terminal module, a cloud service platform, and a doctor terminal module that communicate data via a network.

[0029] The patient terminal module is used to acquire endoscopic examination videos and static images of patients, and after binding the image data with patient identification information, it is encrypted and uploaded to the cloud service platform;

[0030] The cloud service platform includes:

[0031] The endoscopic image time-series database uses the patient's unique identifier as an index and stores all historical and currently uploaded original endoscopic images, intermediate data generated during processing, and result data in a structured manner according to the examination time sequence.

[0032] The intelligent lesion segmentation and registration module is used to receive newly uploaded image data, perform automatic lesion identification, segmentation, quantitative index calculation, and three-dimensional spatial registration between images at different time points;

[0033] The healing dynamic dashboard module is used to acquire the registered image data and the calculated quantitative indicators, and generate a dynamic healing change curve interface for visualization.

[0034] The doctor's terminal module receives and displays the healing dynamic change curve interface, and provides an interactive interface for doctors to input assessment opinions and generate nursing guidance plans, and sends the plans back to the corresponding patient terminal module.

[0035] Furthermore, the intelligent lesion segmentation and registration module includes an intelligent segmentation unit, an index calculation unit, and a spatiotemporal registration unit;

[0036] The intelligent segmentation unit outputs a probability map of a predefined category for each pixel using a pre-trained deep learning segmentation model, thereby generating a lesion segmentation mask.

[0037] The indicator calculation unit receives the lesion segmentation mask image and calculates a set of quantitative indicators for each uniquely identified lesion region in the mask image.

[0038] The spatiotemporal registration unit retrieves historical examination images and feature data from the endoscopic image time series database, and uses the sparse three-dimensional point cloud of the intestine and lens pose extracted with the assistance of the intelligent segmentation unit from the current examination image. Through feature matching and three-dimensional point cloud reconstruction, the current image and historical images are registered in a unified coordinate system.

[0039] Compared with existing technologies, this remote nursing method for gastroenterology patients has the following beneficial effects:

[0040] This invention is based on artificial intelligence and three-dimensional spatial registration for quantitative assessment of mucosal healing. It transforms subjective experience into objective data streams. Through a deep learning segmentation model, it automatically identifies and segments lesion areas in endoscopic images. It reconstructs and registers the endoscopic sequences of the patient's previous examinations to the same three-dimensional intestinal coordinate system. On this unified spatial benchmark, the changes of the first, second, and third quantitative indicators over time are dynamically presented in the form of curves, realizing quantitative tracking of remote monitoring accuracy. This quantitative data based on specific lesion identities allows the adjustment of treatment strategies to be based on small, measurable changes, greatly improving the accuracy of diagnosis and treatment. Doctors can intuitively judge the healing rate and fluctuations and identify the risk of poor treatment response in advance. This intuitive cross-period comparison based on a unified coordinate system, combined with trend analysis, enables the formulation of more targeted and individualized follow-up treatment and monitoring plans.

[0041] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination or study, or may be learned from the practice of the invention. Attached Figure Description

[0042] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0043] Figure 1 A flowchart illustrating the steps of a remote nursing method for patients in the gastroenterology department;

[0044] Figure 2 A diagram showing the modular components of a remote nursing system for patients in the gastroenterology department;

[0045] Figure 3 This is an operation flowchart for a remote nursing system for patients in the gastroenterology department. Detailed Implementation

[0046] To better understand the above technical solutions, a detailed description of the solutions will be provided below in conjunction with the accompanying drawings and specific embodiments. Obviously, the described embodiments are merely 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 creative effort are within the scope of protection of the present invention.

[0047] The terminology used in the embodiments of this invention is for the purpose of describing particular embodiments only and is not intended to limit the invention. The singular forms “a method for remote care of a gastroenterology patient,” “the,” and “the” as used in the embodiments of this invention and the appended claims are also intended to include the plural forms, and “multiple” generally includes at least two unless the context clearly indicates otherwise.

[0048] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that an article or device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such an article or device. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the article or device that includes said element.

[0049] To address the shortcomings of existing technologies, this invention first describes the remote nursing scenario for gastroenterology patients. This invention is primarily applied to the long-term follow-up and precise care of patients with inflammatory bowel disease (IBD), a chronic and recurrent condition where the core treatment goal is to achieve and maintain mucosal healing. In the traditional model, patients frequently travel to the hospital for endoscopic examinations, and doctors rely on subjective experience for assessment, resulting in low efficiency, inconsistent standards, and an inability to quantify and track subtle changes. This invention constructs a system integrating intelligent image analysis, three-dimensional quantitative tracking, and remote visual interaction, aiming to transform mucosal healing assessment from qualitative, subjective, experience-based judgment to quantitative, objective, data-driven decision-making, thereby enabling remote nursing care.

[0050] This invention provides a remote nursing method for patients in the gastroenterology department. Patients upload endoscopic image data, and the cloud service platform uses deep learning and computer vision technology to automatically complete lesion segmentation, three-dimensional registration, and multi-dimensional quantitative index calculation. Finally, the complex time-series data is transformed into intuitive visualization charts through a healing dynamic dashboard, which allows doctors to remotely assess the data and develop personalized nursing plans, forming a closed-loop management system for doctor-patient collaboration.

[0051] Specifically, such as Figure 1 As shown, the large-model-assisted computer vision method for detecting defects in industrial products includes the following steps:

[0052] S100. Construct and maintain an endoscopy image time-series database to store video and image data of a patient’s past endoscopic examinations. Each data is associated with the corresponding patient identity information and examination timestamp.

[0053] S200: Call the deep learning segmentation model to process the endoscopic images in the endoscopic image time series database, automatically segment the lesion area in the image, and register the images at different time points in three-dimensional space to determine the spatial correspondence of the same lesion at different time points.

[0054] S300. For each lesion that has been registered in S200, calculate the quantitative indicator set for each lesion. The quantitative indicator set includes a first quantitative indicator, a second quantitative indicator, and a third quantitative indicator. Associate the quantitative indicator set with the examination timestamp to form a time-series quantitative data sequence of the lesion.

[0055] S400: Based on the generated time-series quantitative data sequence, graphically display the change curves of the first quantitative index, the second quantitative index and the third quantitative index of each lesion over time, and link to display the lesion image comparison view at different time points.

[0056] S500: The doctor, in conjunction with the change curve and the lesion image comparison view, formulates / adjusts the nursing plan and sends it to the patient. The patient provides feedback on the implementation status, and the information is synchronized to the medical staff.

[0057] In its implementation, the S100 constructs and maintains an endoscopy image time-series database. After a patient completes an electronic colonoscopy, during the examination, the endoscopy host, while generating a standard diagnostic report, automatically encrypts and uploads the complete, uncompressed examination video stream and high-definition still images of key frames, along with anonymized patient unique identifiers and examination timestamps, to a designated receiving port on the cloud server. The cloud server's endoscopy image time-series database receives the uploaded data packets. Using the patient's unique identifier as the primary key, the database creates an independent digital folder for each patient. Within this folder, data from each examination is stored as an independent event, strictly sorted by examination timestamp. Each event packet contains: the original video file, extracted keyframe images, and corresponding patient metadata. The database employs a combination of columnar storage and relational database methods, ensuring both efficient access to massive video files and rapid correlation and querying of metadata and subsequent processing results. All data transmission is encrypted and stored. Patient identity information and image data are logically separated and linked through irreversible anonymized IDs, forming a complete access control list and operation log auditing.

[0058] For S200 lesion intelligent segmentation and 3D spatiotemporal registration, the image stream is transformed into structured, spatiotemporally aligned lesion information. When new image data from a recent examination is input, intelligent segmentation is performed using a U-Net deep learning model with an encoder-decoder architecture. The encoder path consists of five convolutional and max-pooling layers, progressively downsampling to extract deep semantic features. The decoder path consists of five deconvolutional and upsampling layers, progressively restoring spatial resolution and fusing shallow features from the same encoder layer via skip connections to improve segmentation boundary accuracy. The final output layer generates a single-channel probability map with the same size as the input image through a 1x1 convolution and a sigmoid activation function. The encoder path feeds a single-frame endoscopic image into the encoder, which consists of convolutional, batch normalization, ReLU activation, and max-pooling units. Each unit extracts image features while halving the spatial size of the feature map and doubling the number of channels. After five downsampling passes, the original image is compressed into a bottleneck feature map containing high-level abstract semantic information. The decoder path is a mirror image of the encoder, consisting of upsampling and convolutional units. Each unit first doubles the size of the feature map, then uses skip connections to concatenate the corresponding high-resolution feature map of the same scale from the encoder's intermediate layers along the channel dimension. This allows the decoder to fuse the detailed information retained by the encoder when restoring the image's spatial resolution, effectively avoiding blurry segmentation results. The decoder output passes through a 1x1 convolutional layer and a sigmoid activation function to generate a probability map of the same size as the input image. Each pixel value in the map represents the probability that it belongs to a lesion region. By setting a threshold, the probabilities are... Figure 2 Values ​​are converted to generate a lesion mask image. All frames are then processed in batches to achieve automatic lesion detection and annotation throughout the entire examination process.

[0059] In practice, simply segmenting the lesions is insufficient for historical comparison, as the angle, distance, and intestinal peristalsis state of the endoscope differ with each examination. Therefore, motion reconstruction technology is used to treat a complete examination video sequence as a series of continuously moving images. Through feature point detection and matching, the trajectory of the endoscope within the intestine is calculated, and a sparse 3D point cloud model of the intestinal wall is simultaneously reconstructed. When processing the patient's Nth examination, the system retrieves the 3D point cloud model generated from the patient's previous examination and used as a baseline from the database. An iterative nearest-point algorithm is employed to find an optimal spatial transformation matrix. This matrix aligns the newly reconstructed point cloud from the Nth examination with the baseline point cloud from the (N-1)th examination in 3D space, minimizing the average distance between corresponding points. This alignment process pastes images taken at different times and angles onto the corresponding positions of the same 3D intestinal model. By transforming the matrix, all lesion masks segmented in the Nth examination are precisely projected from their original two-dimensional image coordinates to a unified three-dimensional intestinal coordinate system. The overlap between new lesions and historical lesions in three-dimensional space is calculated. If the overlap exceeds a set threshold, they are determined to be the same lesion and assigned a unique cross-time point identity ID; otherwise, they are marked as new lesions.

[0060] For each identified lesion, calculate objective and precise numerical characteristics to replace subjective descriptions, including:

[0061] The first quantitative indicator is the three-dimensional projected area: for the registered lesion, obtain all the three-dimensional meshes covered by its mask, and obtain the true projected area of ​​the lesion on the three-dimensional mucosal surface by calculating the sum of the surface areas of these mesh units.

[0062] The second quantitative indicator is the average depth: In the three-dimensional model, an annular region is automatically expanded outside the lesion boundary and defined as a normal mucosal reference area. The average elevation of all three-dimensional points in the lesion area is then calculated. And the average elevation Avg of all points in the reference area. lesion depth This indicator directly quantifies the degree of ulcer depression and serves as a key parameter for assessing healing quality.

[0063] The third quantitative indicator is the vascular texture score: This indicator assesses the microvascular morphology of the mucosa surrounding the lesion, reflecting inflammatory activity and healing trends. Specifically:

[0064] Region Definition and Image Enhancement: A ring-shaped region of interest (ROI) of the mucosa is defined around the lesion after 3D registration. The corresponding original endoscopic image of this region is then subjected to contrast-limited adaptive histogram equalization to highlight vascular texture and suppress uneven illumination.

[0065] Texture feature extraction: Based on the enhanced image, calculate its gray-level co-occurrence matrix. Extract three texture features from this matrix: correlation, entropy, and angular second moment.

[0066] Index synthesis: After normalizing the extracted feature values, the vascular texture index is synthesized using a weighted summation formula. Wherein, VTI is the calculated first vascular texture index, whose value range is normalized to the interval [0, 1]. The higher the value, the closer the vascular texture is to the normal healing state. This represents the average correlation characteristic value, and its value is positively correlated with the regularity of the vascular texture. This represents the inverse measure of texture disorder; its value is positively correlated with texture order. This is the second moment eigenvalue of the mean angle, and its value is positively correlated with the clarity of the blood vessel texture. These are weighting coefficients used to balance the contributions of the three texture features to the final healing state determination result, and VTI is a comprehensive score ranging from 0 to 1, with higher values ​​indicating that the morphology of the mucosal microvessels is closer to a healthy state. All calculated quantitative indicators, along with the examination timestamp, are appended to the time-series data of the lesion and stored in the database.

[0067] The system extracts time-series data of all lesions from the patient's database to obtain registered image data and calculated quantitative indicators. It also generates a dynamic healing curve interface for visualization, displaying the changes of the first, second, and third indicators of the lesions selected by the doctor over time in the form of a line graph. Based on the trend data of all lesions, a text summary is generated.

[0068] Doctors log in through the doctor terminal module and access the healing dynamic dashboard for a designated patient. After reviewing quantitative trends and comparing images, doctors can directly write or select preset templates to generate personalized remote nursing guidance plans. The nursing plans are encrypted and pushed to the corresponding patient terminal module through the system. Patient feedback information is synchronously transmitted back to the cloud database and linked with endoscopic quantitative indicators. When the patient undergoes the next endoscopic examination, the new data will initiate a new round of S200-S400 analysis to update their healing dynamic file. This cycle forms a closed loop of remote nursing that is continuously optimized and driven by objective quantitative data.

[0069] In practical implementation, this method is applicable to a remote nursing system for gastroenterology patients, such as... Figure 2 As shown, the system includes: a patient terminal module, a cloud service platform, and a doctor terminal module;

[0070] The patient terminal module is used to acquire endoscopic examination videos and static images of patients, and after binding the image data with patient identification information, it is encrypted and uploaded to the cloud service platform;

[0071] The cloud service platform includes:

[0072] The endoscopic image time-series database uses the patient's unique identifier as an index and stores all historical and currently uploaded original endoscopic images, intermediate data generated during processing, and result data in a structured manner according to the examination time sequence.

[0073] The intelligent lesion segmentation and registration module is used to receive newly uploaded image data, perform automatic lesion identification, segmentation, quantitative index calculation, and three-dimensional spatial registration between images at different time points;

[0074] The healing dynamic dashboard module is used to acquire the registered image data and the calculated quantitative indicators, and generate a dynamic healing change curve interface for visualization.

[0075] The doctor's terminal module receives and displays the healing dynamic change curve interface, and provides an interactive interface for doctors to input assessment opinions and generate nursing guidance plans, and sends the plans back to the corresponding patient terminal module.

[0076] like Figure 3 As shown, the specific steps of the remote nursing method for gastroenterology patients according to the present invention are as follows:

[0077] (1) Patient endoscopic image database

[0078] Using endoscopic equipment to examine patients, obtain examination videos and capture still images.

[0079] The video and image data from this examination are bound to the patient's unique identifier and examination timestamp to build and maintain a time-series database of endoscopic images.

[0080] The endoscopic image time-series database stores video and image data from each patient's endoscopic examination.

[0081] Each examination data point is linked to the patient's identity information and the examination time.

[0082] (2) Intelligent segmentation of lesion area

[0083] Use pre-trained deep learning models to automatically identify lesion areas in endoscopic images.

[0084] The lesions are segmented at the pixel level to generate a lesion mask image.

[0085] It supports segmentation of various lesion types, such as ulcers, erosions, and edema.

[0086] (3) Three-dimensional spatiotemporal registration

[0087] Endoscopic images at different time points were reconstructed into sparse three-dimensional point clouds.

[0088] The current image and historical images are aligned to the same coordinate system using 3D registration technology.

[0089] Ensure that the spatial location of the same lesion is consistent at different time points.

[0090] (4) Calculate the quantitative indicators of lesions

[0091] Calculate the area of ​​each lesion.

[0092] Calculate the depth of the lesion and the elevation difference between it and the surrounding normal mucosa.

[0093] The score of the vascular texture around the lesion is calculated to reflect the healing status.

[0094] (5) Generate time-series change curves and image comparison views

[0095] By correlating three quantitative indicators of each lesion with time, a time-series data sequence is formed.

[0096] The trend of each indicator over time is displayed in the form of a line graph.

[0097] The system displays a comparison of lesion images at different time points.

[0098] (6) Physician assessment and nursing plan development

[0099] Doctors assess the healing progress of lesions by comparing graphs and images.

[0100] Develop or adjust personalized remote care plans.

[0101] The nursing plan is sent to the patient and feedback on its implementation is received.

[0102] (7) Patient execution and information synchronization

[0103] The patient receives and follows the care plan.

[0104] Patients provide feedback on the implementation status to healthcare professionals.

[0105] The system continuously updates patient data, forming a closed-loop nursing care system.

[0106] In summary, this invention constructs a complete system through S100-S500, from the acquisition of raw endoscopic data to precise lesion quantification and spatiotemporal tracking, and then to data visualization and clinical decision support. It achieves long-term, objective, and multi-dimensional monitoring of the dynamic process of mucosal healing. This intuitive cross-period comparison based on a unified coordinate system, combined with trend analysis, enables physicians to develop more targeted and individualized subsequent treatment and follow-up plans.

[0107] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A remote nursing method for patients in the gastroenterology department, characterized in that, The specific steps of this method are as follows: S100. Construct and maintain an endoscopy image time-series database to store video and image data of a patient’s past endoscopic examinations. Each data is associated with the corresponding patient identity information and examination timestamp. S200: Call the deep learning segmentation model to process the endoscopic images in the endoscopic image time series database, automatically segment the lesion area in the image, and register the images at different time points in three-dimensional space to determine the spatial correspondence of the same lesion at different time points. S300. For each lesion that has been registered in S200, calculate the quantitative indicator set for each lesion. The quantitative indicator set includes a first quantitative indicator, a second quantitative indicator, and a third quantitative indicator. Associate the quantitative indicator set with the examination timestamp to form a time-series quantitative data sequence of the lesion. S400: Based on the generated time-series quantitative data sequence, graphically display the change curves of the first quantitative index, the second quantitative index and the third quantitative index of each lesion over time, and link to display the lesion image comparison view at different time points. S500: The doctor, in conjunction with the change curve and the lesion image comparison view, formulates / adjusts the nursing plan and sends it to the patient. The patient provides feedback on the implementation status, and the information is synchronized to the medical staff.

2. The remote nursing method for gastroenterology patients according to claim 1, characterized in that, In S200, the deep learning segmentation model is a U-Net network based on an encoder-decoder architecture, which consists of: The encoder path consists of multiple convolutional layers and max pooling layers, used to extract feature maps layer by layer from the input endoscopic images and compress the features to capture the contextual semantic information of the lesions. Decoder path: Consists of multiple deconvolutional layers and upsampling layers, used to restore the feature maps extracted by the encoder to the spatial resolution of the original input image, and perform pixel-level localization; Final output layer: The features output by the decoder path are mapped to a single-channel probability map of the same size as the input image. The value of each pixel in the single-channel probability map represents the probability that it belongs to the lesion region.

3. The remote nursing method for gastroenterology patients according to claim 2, characterized in that, In step S200, the specific steps for automatically segmenting the lesion region in the image are as follows: The endoscopic images to be processed are input into the deep learning segmentation model; The encoder path of the deep learning segmentation model extracts and downsamples the input endoscopic image to obtain image features containing high-level semantic information. The decoder path upsamples and refines image features, and restores spatial accuracy by fusing features of the same scale from the encoder path. The final output layer generates a probability map. By setting a probability threshold, pixels with probability values ​​higher than the probability threshold in the probability map are identified as lesion pixels, thereby generating a binarized lesion mask image.

4. The remote nursing method for gastroenterology patients according to claim 1, characterized in that, In step S200, images from different time points are registered in three-dimensional space. The registration process is as follows: The endoscopic video sequence was reconstructed to reconstruct a sparse three-dimensional point cloud of the intestine and the camera pose during the examination. The 3D point cloud reconstructed in the current inspection is registered with the historical 3D point cloud used as a reference through the iterative nearest point algorithm to find the spatial transformation matrix that minimizes the average distance between the two point clouds. Using the spatial transformation matrix, all currently examined image data and their corresponding lesion segmentation results are uniformly mapped to the historical three-dimensional coordinate system of the reference.

5. A remote nursing method for gastroenterology patients according to claim 1, characterized in that, In S300, the first quantification index is the area of ​​the lesion. On the completed three-dimensional registration, the set of three-dimensional mesh vertices covered by the lesion mask image is obtained. By calculating the sum of the surface areas of the three-dimensional mesh, the actual projected area of ​​the lesion in three-dimensional space is obtained.

6. A remote nursing method for gastroenterology patients according to claim 1, characterized in that, In step S300, the second quantification indicator is the depth of the lesion. Based on the completed 3D registration, the 3D point cloud data of the lesion region is extracted. Extracting three-dimensional point cloud data of normal mucosal areas within a preset range around the lesion. ,pass The depth of the lesion was calculated. Among them, Avg Point cloud representing normal mucosal area The average of the elevation values ​​of all points Point cloud representing the lesion area The average elevation value of all points.

7. A remote nursing method for gastroenterology patients according to claim 1, characterized in that, In step S300, the third quantification index is the vascular texture score of the lesion. Based on the completed 3D registration, using the segmented lesion boundary as a reference, a pre-defined annular region is expanded as the mucosal region of interest (ROI) for vascular texture analysis. The ROI undergoes contrast-limited adaptive histogram equalization. Based on the image after histogram equalization, the gray-level co-occurrence matrix of the ROI is calculated. This gray-level co-occurrence matrix describes the image's gray levels. The pixels and grayscale are The probability of pixels co-occurring is used to extract texture features from the gray-level co-occurrence matrix, including correlation, entropy, and second moment of angle. The average value of the texture features is extracted as a comprehensive feature quantity, and the comprehensive feature quantity is synthesized into a single first vascular texture index through linear weighting.

8. A remote nursing system for gastroenterology patients, applicable to the remote nursing method for gastroenterology patients as described in any one of claims 1-7, characterized in that, The system includes a patient terminal module that communicates data via a network, a cloud service platform, and a doctor terminal module; The patient terminal module is used to acquire endoscopic examination videos and static images of patients, and after binding the image data with patient identification information, it is encrypted and uploaded to the cloud service platform; The cloud service platform includes: The endoscopic image time-series database uses the patient's unique identifier as an index and stores all historical and currently uploaded original endoscopic images, intermediate data generated during processing, and result data in a structured manner according to the examination time sequence. The intelligent lesion segmentation and registration module is used to receive newly uploaded image data, perform automatic lesion identification, segmentation, quantitative index calculation, and three-dimensional spatial registration between images at different time points; The healing dynamic dashboard module is used to acquire the registered image data and the calculated quantitative indicators, and generate a dynamic healing change curve interface for visualization. The doctor's terminal module receives and displays the healing dynamic change curve interface, and provides an interactive interface for doctors to input assessment opinions and generate nursing guidance plans, and sends the plans back to the corresponding patient terminal module.

9. A remote nursing system for gastroenterology patients according to claim 8, characterized in that, The intelligent lesion segmentation and registration module includes an intelligent segmentation unit, an index calculation unit, and a spatiotemporal registration unit. The intelligent segmentation unit outputs a probability map of a predefined category for each pixel using a pre-trained deep learning segmentation model, thereby generating a lesion segmentation mask. The indicator calculation unit receives the lesion segmentation mask image and calculates a set of quantitative indicators for each uniquely identified lesion region in the mask image. The spatiotemporal registration unit retrieves historical examination images and feature data from the endoscopic image time series database, and uses the sparse three-dimensional point cloud of the intestine and lens pose extracted with the assistance of the intelligent segmentation unit from the current examination image. Through feature matching and three-dimensional point cloud reconstruction, the current image and historical images are registered in a unified coordinate system.

Images