Intelligent analysis system based on gastrointestinal concurrent inflammation

Abstract

The invention discloses an intelligent analysis system based on digestive tract concurrent inflammation, which comprises a data acquisition layer, a data processing layer, an intelligent analysis layer, a decision support layer and a user interaction layer, and is characterized in that the data acquisition layer is responsible for collecting data from various channels; the data processing layer is responsible for cleaning, integrating and standardizing the original data collected by the data acquisition layer, so that the subsequent intelligent analysis layer can perform further analysis and mining; the intelligent analysis layer is used for deeply analyzing and mining the data cleaned, integrated and standardized by the data processing layer so as to identify potential inducements of digestive tract inflammation; a wireless capsule endoscope image ulcer detection model based on curvelet is adopted; the decision support layer is responsible for converting an analysis result generated by the intelligent analysis layer into suggestions and information which have guiding significance to clinical decisions; the user interaction layer provides an intuitive and easy-to-use user interface, so that doctors and other medical personnel can interact with the system conveniently.

Description

Technical Field

[0001] The present invention relates to an evaluation system, in particular to an intelligent analysis system based on digestive tract inflammation complications. Background Art

[0002] The incidence of inflammatory diseases of the digestive tract is on the rise worldwide, especially the incidence of inflammatory bowel disease (such as Crohn's disease and ulcerative colitis). This reflects the impact of changes in modern lifestyle, eating habits and environmental factors on digestive tract health. The occurrence of digestive tract complications is related to a variety of factors, including poor eating habits, life stress, pathogen infection (such as Helicobacter pylori), genetic factors, autoimmune abnormalities, etc. The complexity of these factors increases the difficulty of prevention and treatment. The clinical manifestations of digestive tract inflammation are diverse, including diarrhea, abdominal pain, fever, loss of appetite, etc. For inflammatory bowel disease, serious complications such as intestinal stenosis, obstruction, and perforation may also occur, which seriously affects the quality of life of patients. Methods for treating digestive tract inflammation include drug therapy, endoscopic therapy, surgical treatment, and nutritional support. However, due to the heterogeneity of the disease and individual differences, the treatment effect often varies from person to person. In addition, long-term treatment may bring drug side effects and economic burden. Preventing the occurrence of digestive tract inflammation requires improving living habits, eating habits, and strengthening personal hygiene. However, the public's attention to digestive tract health and related knowledge level still need to be improved. The early symptoms of gastrointestinal inflammation may be atypical and easily confused with other diseases, making early diagnosis difficult. This may delay treatment and increase the risk of complications. Although the treatment principles have been clarified, the degree of individualization of treatment plans still needs to be improved due to the heterogeneity of the disease and individual differences. Current treatment plans are often based on experience or clinical trial results, which are difficult to fully meet the needs of each patient. Patients' lack of understanding and attention to gastrointestinal inflammation leads to poor treatment compliance. Some patients may stop taking medication or ignore treatment recommendations due to symptom relief, which affects the treatment effect and prognosis. Gastrointestinal inflammatory diseases often require long-term treatment and management. However, long-term treatment may bring drug side effects and economic burden, and the patient's quality of life may also be affected. Therefore, how to balance the treatment effect with the patient's quality of life has become a challenge. Although some progress has been made in the research of gastrointestinal inflammatory diseases, there are still many unknown areas to explore. Insufficient research investment and innovation may limit the development and application of new treatment methods and prevention strategies. In summary, the current status of gastrointestinal complications of inflammation shows that we still need to make continuous efforts in prevention, diagnosis and treatment. By increasing research investment, improving early diagnosis rates, optimizing treatment options, and strengthening patient education and management, we can better meet this challenge and improve patients' quality of life and health. Summary of the invention

[0003] In order to overcome the deficiencies of the prior art solutions, the present invention provides an intelligent analysis system based on digestive tract complications and inflammation.

[0004] In order to achieve the above purpose, the technical solution of the present invention is: the intelligent analysis system based on digestive tract inflammation includes five main parts: data acquisition layer, data processing layer, intelligent analysis layer, decision support layer and user interaction layer, among which,

[0005] 1. Data collection layer, responsible for collecting data from various channels, including hospital information systems, laboratory information systems, and image acquisition systems.

[0006] Image acquisition systems, including wireless capsule endoscopes, ultrasonic wireless capsules, CT, and MRI imaging equipment, are used to collect images of the digestive tract.

[0007] The specific components of wireless capsule endoscope are as follows:

[0008] Housing: The housing of the wireless capsule endoscope is made of materials that meet biocompatibility standards, such as medical-grade plastics or metal alloys. This housing not only protects the internal components from the external environment, but also ensures the safe passage of the capsule in the digestive tract. The housing design takes into account the patient's swallowing comfort and the stability of the capsule in the digestive tract.

[0009] Camera: The camera is one of the core components of the wireless capsule endoscope, responsible for capturing images of the inside of the digestive tract. It is a miniature camera with high resolution and wide-angle field of view, which can clearly capture the fine structure of the digestive tract mucosa. The camera is combined with an optical system (such as a lens group) to ensure the quality and clarity of the image. Specifically, the camera is a metal oxide semiconductor imaging chip camera.

[0010] Light source: In order to illuminate the inside of the digestive tract, the wireless capsule endoscope has a built-in light source. This light source is an LED lamp or other low-power, high-brightness light source. The design of the light source needs to take into account the uniformity and brightness of the lighting to ensure that the camera can capture a clear image. Specifically, 6 white light-emitting diode lighting sources are used.

[0011] Image sensor: The image sensor is responsible for converting the light signal captured by the camera into an electrical signal and further into digital image data. This sensor is a highly sensitive CMOS or CCD sensor that can capture high-quality images.

[0012] Wireless communication module: The wireless communication module is a key component for the wireless capsule endoscope to communicate with external devices (such as data receiving devices). It includes a radio frequency transmitter and an antenna element, which can wirelessly transmit image data to the outside of the body.

[0013] Power system: The power system of the wireless capsule endoscope consists of batteries, which provide power for the camera, light source, image sensor, and wireless communication module components. The battery needs to have sufficient capacity and stability to ensure the working time of the capsule in the digestive tract. Specifically, the power source is 2 silver oxide batteries.

[0014] Control unit: Although wireless capsule endoscopes do not require a complex control unit, some advanced models contain a microprocessor or other control elements to control the camera's shooting parameters, the brightness of the light source, the rate of wireless communication, etc.

[0015] Other auxiliary components: The wireless capsule endoscope also contains other auxiliary components, such as temperature sensors, pressure sensors, etc., which are used to monitor the environmental parameters inside the digestive tract. These components provide additional diagnostic information to help doctors understand the patient's digestive tract condition more comprehensively.

[0016] 2. The data processing layer is responsible for cleaning, integrating, and standardizing the raw data collected by the data collection layer so that the subsequent intelligent analysis layer can conduct further analysis and mining. It includes the following functions:

[0017] Data cleaning: During the data collection process, the original data needs to be cleaned to remove duplicate, erroneous or irrelevant information to ensure the quality of the data.

[0018] Data integration: Integrate data from different systems to form a structured data set.

[0019] Feature extraction: Extract key features related to gastrointestinal inflammation from the integrated data, such as inflammatory marker levels, imaging features, etc.

[0020] Data standardization: Standardize the data to facilitate subsequent intelligent analysis.

[0021] Among them, data cleaning was performed using a non-information frame system for automatic detection of enteroscopy for digestive tract complications;

[0022] An effective computer-assisted system for automatically detecting and estimating disease features present in enteroscopy for digestive tract complications promotes standardized disease assessment. An important obstacle to automatically analyzing enteroscopy videos for digestive tract complications is the large number of non-informative frames caused by motion blur, blurred vision caused by debris, changes in light intensity and image. Those non-informative frames interrupt disease severity estimation by providing non-informative or conflicting information. An effective classification method is crucial to discard non-informative frames while retaining informative frames. The automatic detection system for enteroscopy for digestive tract complications has the following functions:

[0023] 2.1. Collect videos

[0024] Images of the digestive tract of patients with inflammatory bowel disease were collected. Videos were recorded at 30 frames per second at 1920×1080 resolution. Video frames sampled at a rate of one frame per second were manually annotated by a board-certified gastroenterologist. Specifically, the following four types of frames were annotated as non-informative:

[0025] A motion blurred frame is sufficient to blur the colon vasculature.

[0026] BFrames captured when the camera view is obscured by excessive liquid or solid debris.

[0027] C Frame captured when the camera was too close to the colon wall.

[0028] D The picture is overexposed due to a failure of automatic lighting control.

[0029] The frame distribution in each digestive tract inflammatory complication enteroscopic examination video was used to identify and remove the above four types of non-informative frames.

[0030] 2.2. Preprocessing

[0031] Frames from different digestive tract inflammatory complication enteroscopy examinations were preprocessed to ensure consistency. First, each frame was binarized, the largest component in the frame was identified by the content captured by the camera, and the smallest rectangular area containing the largest component was used to crop the original frame. After that, zero padding was added to fill the rectangular area into a square area, and then the size of the square area was resized to 256×256 pixels. The region of interest for feature extraction was an octagon.

[0032] 2.3. Bottleneck feature acquisition

[0033] Bottleneck features are used for non-informative frame classification. More than 10,000 annotated frames were used for model training. To avoid bias from individual patient characteristics, a pre-trained model was used and fine-tuned using annotated training data. The activation map from the last average pooling layer was considered as the bottleneck feature. The number of bottleneck features was 2048.

[0034] Feature extraction

[0035] The hue, saturation, and value channels in the HSV color space correspond to color, grayscale, and brightness, respectively. The hue channel is visualized by a color map. Unlike the RGB color space, the HSV color space separates color from intensity. Extracting features from frames in the HSV color space provides additional information. While many pre-trained models are available for generating features in the RGB color space, large-scale datasets suitable for HSV feature extraction are limited, where reflection, shadow, and brightness in an image should be relevant for classification tasks.

[0036] In order to overcome the shortcomings of large-scale datasets, the frames are converted from RGB to HSV, and a feature extraction algorithm is proposed. Features are extracted to characterize the frames in the digestive tract inflammatory colonoscopy video. The details of each feature extraction are as follows:

[0037] 2.4.1 Specular reflection mask establishment: Due to specular reflections from the surface of the colon, many frames contain many bright areas, which will significantly increase the number of edges and contrast of the image. To eliminate the effect of these reflections, a reflection mask is established for each frame using threshold and dilation operations on the saturated channel. The fraction of reflective areas in the region of interest is calculated as a feature. The reflection mask is used to correct features in other categories.

[0038] 2.4.2. Intensity Statistics: After removing the pixels inside the reflective mask and outside the region of interest, the mean, variance, skewness and kurtosis of the pixel intensities in the hue, saturation and value channels are calculated respectively. The mean gives the average level, while the variance calculates the heterogeneity. In addition, the skewness and kurtosis measure the asymmetry and tail of the distribution respectively. The intensity statistics characterize the distribution of color components, grayscale and brightness in the region of interest.

[0039] 2.4.3 Edge feature extraction: First, contrast-limited adaptive histogram equalization is used to enhance the contrast in the value channel, where the contrast transformation function is calculated on the local area. Secondly, the Canny edge detector is used to generate an initial edge mask. To eliminate the effect of high reflection, the edges within the reflection mask are removed. In addition, each edge component is analyzed, and edges with an eccentricity less than 0.9 are removed. The percentage of edges in the region of interest is calculated as an edge feature. The Hough transform is used to find lines in the edge. Considering that a typical information frame contains a clear view of the colon lumen and colon folds, the number of linear edges is also used as an edge feature. The fold contours of the colon are successfully detected. Edge detection using the Canny method will also find sharp intensity gradients in the colon wall (e.g., blood vessels in the colon wall), but line detection responses tend to gather at fold contours and significant disease features.

[0040] 2.4.4 Gray-level co-occurrence matrix calculation: In order to eliminate the influence of the boundary of the region of interest, inward interpolation is used to fill the area outside the region of interest. Similarly, in order to eliminate the influence of reflection, the area falling within the reflection mask is also filled. The gray-level co-occurrence matrix is ​​calculated for each channel separately. The pixel intensity is quantized into 32 levels. The second-order statistics of the gray-level co-occurrence matrix include contrast F con Energy F ener and uniformity F hom , calculated as follows:

[0041]

[0042] v(i,j) represents the gray level co-occurrence matrix value at the position (i,j).

[0043] 2.4.5 Blur calculation: The degree of blur is an important indicator for the classification of uninformative images. To measure blur, rectangular regions (sub-regions) obtained from inward interpolation are used. A Gaussian filter is used to generate an absolute difference map between the original and blurred sub-regions. The mean and standard deviation of the differences are calculated as blur metrics. This is based on the assumption that the Gaussian filter has a small effect on blurred images. In addition, the reference-free image quality measure of blurred images in the frequency domain proposed in is calculated. Since blurred images are also caused by optical defocus, a set of focus metrics are calculated. By convolving the discrete Laplacian mask with the focus measurement of the input image, the measurement value F LAPE Written as:

[0044]

[0045] Where I represents the input image, H and W represent the height and width of the input image respectively.

[0046] By dividing the image into multiple square blocks and calculating the energy ratio F of the AC and DC coefficients of each block in the discrete cosine transform domain DCTR , used to measure focus:

[0047]

[0048] Where S is the block size. u,v is the square of the AC coefficient in block (u, v), EDC u,v is the square of the DC coefficient in block (u, v). Preferably, S=15 is used.

[0049] 2.5. Classifier Training

[0050] A combination of handcrafted features and bottleneck features was used to train the random forest model. Considering the large number of features after feature fusion, the random forest model with automatic feature selection was chosen.

[0051] In the digestive tract inflammatory colonoscopy video, image clarity is affected by camera motion. In addition, image brightness and clarity are affected by variable characteristics of the internal colon environment, including the amount of water or debris, surface texture, and the distance between the camera and the colon wall. Therefore, image analysis in the HSV color model (separating color components from intensity) provides additional information. A set of hand-crafted features were extracted in the HSV color space, including measures of edges, reflections, blur, and focus. The hand-crafted features in the HSV color space and the RGB color space effectively improved the classification performance.

[0052] Data integration integrates data from different data sources and different formats to form a unified data set. Convert data in different formats (such as text, images, videos, etc.) into a unified format for subsequent processing. Use data integration technology (such as ETL tools, data warehouses, etc.) to merge data from different data sources and build a unified data view. Associate different data tables or data sets through key fields (such as patient ID, examination time, etc.) to achieve data interconnection.

[0053] Data standardization ensures data consistency and comparability, and provides a standardized data foundation for subsequent intelligent analysis. Classified data is uniformly coded, such as disease names and drug names, using internationally accepted coding standards (such as ICD, ATC, etc.). Numerical data is uniformly processed, such as length units converted to meters, weight units converted to kilograms, etc. For data that requires numerical calculations (such as age, laboratory indicators, etc.), normalization or standardization is performed to eliminate the impact of dimensional differences on analysis results.

[0054] 3. The intelligent analysis layer uses artificial intelligence technologies such as machine learning and deep learning to conduct in-depth analysis and mining of the data that has been cleaned, integrated, and standardized by the data processing layer to identify potential causes of gastrointestinal inflammation. It uses a curvelet-based wireless capsule endoscope image ulcer detection model, which is a multi-resolution analysis tool. Its performance is better than the wavelet analysis model in the two-dimensional domain. The selected color space is YCbCr. The image is first decomposed into multiple sub-bands of different scales and directions. Then, the porosity index of the selected sub-band is calculated and a feature vector is formed. The curvelet-based wireless capsule endoscope image ulcer detection model has the following functions 3.1. Curvelet transform

[0055] The curvelet transform overcomes the weakness of wavelets and effectively represents two-dimensional singular points in images, such as lines and curves, which often appear in medical images. -j , direction θ l , position (j, l, k) and current time t, the moment when the image is at position (j, l, k) By translating and rotating the original curve The definitions are as follows:

[0056]

[0057] in, is the rotation θ l radians. The variable θ l is the rotation angle An equally spaced sequence of l ≤2π; is the sequence parameter of the translation. through its Fourier transform is defined and represented in polar coordinates (r, θ) in the Fourier domain as:

[0058]

[0059] The support of is a polar parabolic wedge defined by the supports of W() and V() (radial and angular windows) respectively, with a scale-dependent window width applied in each direction; W() and V() satisfy the partition of unity property. In continuous frequency v, the curvelet coefficient α of the two-dimensional function f j,l,k is defined as the inner product:

[0060]

[0061] 3.2. Differential porosity analysis

[0062] The porosity is calculated using a sliding box algorithm, which is applicable to binary datasets, and a differential porosity applicable to grayscale image analysis is introduced. The differential porosity is calculated by a differential box-counting algorithm, which uses a sliding box r of size r×r pixels and a sliding window w of size w×w pixels (r < w). The window scans the entire image, while the box scans the pixels bounded by the window in order to calculate the box quality at each position of the window. Depending on the pixel values contained in the box, more than one column of cubes r×r×r is required to cover the image intensity surface. Numbers 1, 2, etc. are assigned to the cubes from bottom to top, and the height difference n(i, j) of the columns is calculated. When the box slides over the window, the sum of all n(i, j) is the box quality of the window. Then, the differential porosity value L(r, w) is calculated through the same procedure as in the sliding box algorithm.

[0063] 3.3. Distinguishing tissues

[0064] Normal and concurrent inflammatory tissues in wireless capsule endoscopy images are distinguished by developing a texture-based detection and classification scheme. The image is decomposed into color channels, where Y is the luminance component and Cb and Cr are the blue-difference and red-difference chrominance components. The Y, Cb, Cr channels are obtained from the R, G, B channels as follows:

[0065] Y = 16 + (66R + 129G + 25B)

[0066] Cb = 128 + (-38R - 74G + 112·B)

[0067] Cr = 128 + (112R - 94G - 18B)

[0068] In the next step, a discrete curvelet transform is applied to each color channel by a wrapping method. Four scale decompositions are used. In addition, each scale contains a certain number of angles, which vary from scale to scale. 8 angles (must be a multiple of 4, with a minimum number of 8) are selected as the second scale. The use of 16 angles for the second scale was also tested. However, it was noted that information is repeated, which will lead to increased computational time and data redundancy. Specifically, when 16 angles are used for the second scale, the first 8 sub-images are very similar to the last 8 sub-images. The above choices of scales and angles result in 1, 8, 16, 16 angles per scale, respectively, for a total of 41 discrete curvelet transformed sub-images.

[0069] The next step consists in calculating the differential porosity index for each subband. The differential porosity is calculated for a constant frame size r = 3 pixels and a window size w = 4frame_max, where frame_max is equal to the minimum size of each subimage. A differential porosity curve is obtained for each subimage with as many coefficients as possible. Each differential porosity curve is normalized to the highest value to ensure the same reference level. The differential porosity curve is modeled using a hyperbolic function Λ(w), defined as follows

[0070]

[0071] The parameters (a, b, c) are independent variables calculated as the solution of the least squares problem. These three parameters represent the differential porosity-based texture features of the wireless capsule endoscopy images. Six common statistical features, namely, mean, standard deviation, entropy, energy, skewness, and kurtosis, are also extracted from the differential porosity curves.

[0072] As mentioned above, for each set A, B or C, 41 different eigenvectors are obtained, one for each of the 41 different angles (subbands) of the discrete curvelet transform. The purpose of this feature extraction method is to test the discrimination ability of each angle individually.

[0073] 4. Decision support layer

[0074] The decision support layer of the intelligent analysis system based on digestive tract complications is a key component of the system. It is mainly responsible for converting the analysis results generated by the intelligent analysis layer into suggestions and information that are instructive for clinical decision-making. It includes the following functions:

[0075] 4.1 Image and video data annotation,

[0076] Image annotation was performed for images and videos from patients who underwent full laparoscopy and obtained wireless capsule images of all 5 intestinal segments, including clear white-light endoscopic images without image-enhanced endoscopy, feces, blur or halo. All images were evaluated by 4 wireless capsule physicians with 30, 11, 4 and 6 years of experience, respectively.

[0077] 4.2 Artificial Intelligence Algorithms,

[0078] The AI ​​architecture is divided into the following modules: image classification framework, video processing pipeline, and weighted evaluation system. Due to its reliability and effectiveness, the pre-trained convolutional neural network framework was selected as the main one. It is a feature extractor, the first step of the training protocol, and then a probability score ranging from 0 to 1 is created for each image by the trained CNN. A dedicated CNN model is designed for the video processing pipeline, which contains a visual clarity module and an image similarity determination module. This CNN model is used to pre-process highly complex data and obtain image sequences as input to the evaluation module. Therefore, the visual clarity model first detects all frames in the visual clarity stage to remove low-definition images. The similarity detection process judges the similarity of each frame with 4 adjacent images, and then removes images that are too similar to the previous and next frames to prevent these images from being repeatedly evaluated.

[0079] 4.3 Evaluation model development

[0080] The evaluation system consists of two parts: baseline evaluation model and weighted evaluation model.

[0081] 4.3.1 Baseline Evaluation Model

[0082] The baseline assessment model was used to determine a quick sketch of the full-length intestinal tract, which was expressed as follows:

[0083]

[0084] Among them, M i (0-3) is an approximate clinician estimate reflecting the severity of inflammation associated with a particular bowel segment. Length length and N are the length of the inflamed portion and the number of inflamed bowel segments (0-5). The values ​​obtained using this equation are determined subjectively by each endoscopist based on their experience and have considerable variability.

[0085] 4.3.2 Weighted Evaluation Model

[0086] A weighted evaluation model was established and applied to the image sequence. Based on clinical experience, the intestine was divided into a fixed number of regions, including 20 images in the cecum region, 20 images in the transverse colon region, 20 images in the descending colon region, 15 images in the sigmoid colon region, and 10 images in the rectum region. Each region contains a certain number of intestinal images, and the system calculates the area score Arsco according to the following formula:

[0087] Arsco=(Num mild PPV mild +2Num erate PPV erate +3Num severe PPV severe )

[0088] Among them, avg is the average value of the evaluation prediction of all endoscopic images for the image sequence, Numx represents the proportion of the evaluation results in the intestinal area, and PPVx (positive predictive value) is the statistical analysis value predicted by the AI ​​algorithm of the Mayo classification task, which measures the ratio of true positive cases to classified positive cases. Clinicians pay attention to highly severe inflammation during clinical diagnosis. In the calculation formula of area score, mild, erate and severe represent mild, moderate and severe, respectively.

[0089] After determining the score of each region of the internal segment, the segmental Segsco score is generated by the following linear combination:

[0090]

[0091] Among them, ratio and L represent the percentage of each inflammation severity and the number of image sequences of the intestinal segment, respectively.

[0092] 4.4 Statistical analysis

[0093] The primary outcome measure was defined as the accuracy of the AI ​​algorithm's predictions. The harmonic mean of the positive predictive value and the negative predictive value was chosen as the basis for evaluating the model's performance. The secondary outcome measures were the outputs of the evaluation system, including the results of the area assessment and the segment assessment. The distribution of values ​​for the area score was discussed. In addition, to verify the validity of the segment score, the validation part focused on the correlation between the weighted evaluation system and the observer's baseline assessment model. The agreement was evaluated by confusion matrix and other metrics, including positive predictive value, negative predictive value, sensitivity, and specificity.

[0094] 5. User Interaction Layer

[0095] Provide an intuitive and easy-to-use user interface to facilitate doctors and other medical personnel to interact with the system. Display analysis results and diagnostic suggestions in the form of charts, reports, etc. to help doctors quickly understand the condition and develop treatment plans. Support doctors to enter patient information, query historical records, modify diagnostic suggestions, and other operations to improve work efficiency. The user interaction layer of the intelligent analysis system based on gastrointestinal complication inflammation is a key part of information exchange and interaction between the system and users. It is mainly responsible for receiving user input, displaying the system's analysis results, and providing a user-friendly operation interface. The following is the detailed composition of the user interaction layer:

[0096] 5.1 User Interface Design

[0097] Intuitive: Use a simple and clear interface design to ensure that users can easily understand and operate it.

[0098] Responsiveness: Ensure that the system interface can quickly respond to user operations and improve user experience.

[0099] Customizability: Provide personalized setting options, allowing users to adjust the interface layout, font size, etc. according to their needs.

[0100] 5.2 Input Module

[0101] Data input: Support multiple data input methods, such as text input, file upload, voice input, etc. Provide clear data input guidance to ensure that users can enter the required information correctly and completely.

[0102] Patient Information Management: Allows users to create, edit and query basic patient information, including name, age, gender, and medical history.

[0103] 5.3 Result display and feedback

[0104] Analysis result display: The results generated by the intelligent analysis layer are displayed to users in an intuitive manner.

[0105] Detailed explanation: Provide detailed explanation and description of the analysis results to help users understand the cause, pathological process and treatment plan.

[0106] Interactive feedback: Allow users to provide feedback on analysis results so that the system can be continuously optimized.

[0107] 5.4 Auxiliary tools and resources

[0108] Clinical Guidelines and Knowledge Base: Provide clinical guidelines and medical knowledge base resources related to gastrointestinal inflammation for users' reference and learning.

[0109] Case database: displays typical cases and successful cases to help users understand the treatment methods and effects of different diseases.

[0110] Online help and customer service: Provide online help documents, FAQ, and customer service support to solve problems encountered by users during use.

[0111] 5.5 Security and Privacy Protection

[0112] Data encryption: Encrypt the data input by the user and the information during transmission to ensure data security.

[0113] Privacy protection: Strictly abide by relevant laws and regulations to protect users' personal privacy information from being leaked.

[0114] Beneficial effects:

[0115] (1) The intelligent analysis system for the causes of gastrointestinal inflammatory complications is a complex medical information system that aims to intelligently identify and analyze the causes of inflammation by collecting, processing and analyzing data related to gastrointestinal inflammation, providing strong support for clinical diagnosis and treatment;

[0116] (2) The data collection layer collects data related to gastrointestinal inflammation through various channels and methods, and ensures the quality and accuracy of the data through data quality control measures. These data will provide a solid foundation for subsequent data processing and analysis.

[0117] (3) The data processing layer provides high-quality and standardized input data for the intelligent analysis layer through steps such as data cleaning, data integration, data standardization and data preprocessing, which is an important guarantee for ensuring the accuracy and reliability of the system analysis results. The proposed automatic detection system for non-information frames used for colonoscopy is essential for further analysis of colonoscopy videos. Accurate detection and removal of non-information frames can effectively improve the accuracy of disease severity estimation and reduce computational costs.

[0118] (4) The intelligent analysis layer provides doctors with accurate diagnostic support and personalized treatment plan recommendations by selecting appropriate algorithms, building efficient models, and realizing multi-source data fusion analysis and association rule mining. The existence of this layer greatly improves the accuracy and efficiency of identifying the causes of gastrointestinal inflammation, providing strong support for clinical diagnosis and treatment. A curvelet-based wireless capsule endoscope image ulcer detection method is proposed. The performance of the discrete curvelet transform-differential porosity scheme is compared with the results obtained after the application of the proposed method. The popular curvelet-based LBP analysis was performed on the same data set, showing 82.2% acc, 87.1% sensitivity and 77% specification, which is 4.34% and 7.51% lower acc. and specification, these results show that the extraction of texture features based on differential porosity of discrete curvelet transform subbands is more effective than similar methods in ulcer detection.

[0119] (5) The decision support layer is mainly responsible for converting the analysis results generated by the intelligent analysis layer into suggestions and information that are instructive for clinical decision-making. A new deep learning-based scoring system was established to evaluate the endoscopic images of UC patients. The system also accurately describes the severity and distribution of inflammatory activity through full-length intestinal endoscopic videos. BRIEF DESCRIPTION OF THE DRAWINGS

[0120] Figure 1 The block diagram of the intelligent analysis system based on digestive tract inflammation of the present invention; DETAILED DESCRIPTION

[0121] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0122] like Figure 1 As shown, the intelligent analysis system based on digestive tract inflammation,

[0123] The intelligent analysis system based on digestive tract inflammation is a complex medical information system that aims to provide strong support for clinical diagnosis and treatment by collecting, processing and analyzing data related to digestive tract inflammation, intelligently identifying and analyzing the causes of inflammation. The following is a detailed description of the system:

[0124] The intelligent analysis system based on digestive tract complications includes five main parts: data collection layer, data processing layer, intelligent analysis layer, decision support layer and user interaction layer.

[0125] 1. Data collection layer, which is responsible for collecting data from various channels, including patients' clinical information (such as symptom description, medical history, physical examination results, etc.), laboratory test results (such as blood tests, urine tests, stool tests, etc.), imaging data (such as gastroscopy, colonoscopy, ultrasound wireless capsule images, etc.), and genetic and environmental factor data, including hospital information systems, laboratory information systems, and image acquisition systems.

[0126] Hospital information systems are used to collect: Clinical information: including patient symptom descriptions, medical history, physical examination results, etc. Diagnostic information includes the doctor's diagnosis and preliminary judgment of the condition. Treatment information includes the patient's treatment plan, medication records, and treatment effect feedback.

[0127] The laboratory information system is used for biochemical tests: such as the results of blood tests (routine blood tests, liver function, kidney function, etc.), urine tests (routine urine tests, urine sediment analysis, etc.), and stool tests (occult blood tests, parasite egg tests, etc.). Microbiological tests: such as the results of Helicobacter pylori tests and virus tests. Use patient questionnaires to survey lifestyle habits: such as eating habits, drinking and smoking history, exercise status, etc. Environmental factors: such as living environment, occupational exposure, etc. Family medical history: understand whether there is a genetic history of digestive tract diseases in the patient's family. Biological sample collection: Blood samples: used to detect inflammatory markers, genetic genes, etc. Urine samples: used to detect related indicators in urine. Stool samples: used to detect parasite eggs, intestinal flora, etc.

[0128] Image acquisition systems, including wireless capsule endoscopes, ultrasonic wireless capsules, CT, and MRI imaging equipment, are used to collect images of the digestive tract.

[0129] Among them, the wireless capsule endoscope is an advanced medical device used to non-invasively examine the inside of the human digestive tract. This capsule is only 11 mm × 26 mm and weighs 3.7 grams. After a night of fasting, the patient swallows the capsule, which moves forward passively through physical peristalsis. The capsule captures a series of frames at a rate of two frames per second for the duration of the battery life (8 hours). The total number of images generated during the inspection is about 55,000.

[0130] The specific components of wireless capsule endoscope are as follows:

[0131] Housing: The housing of the wireless capsule endoscope is made of materials that meet biocompatibility standards, such as medical-grade plastics or metal alloys. This housing not only protects the internal components from the external environment, but also ensures the safe passage of the capsule in the digestive tract. The housing design takes into account the patient's swallowing comfort and the stability of the capsule in the digestive tract.

[0132] Camera: The camera is one of the core components of the wireless capsule endoscope, responsible for capturing images of the inside of the digestive tract. It is a miniature camera with high resolution and wide-angle field of view, which can clearly capture the fine structure of the digestive tract mucosa. The camera is combined with an optical system (such as a lens group) to ensure the quality and clarity of the image. Specifically, the camera is a metal oxide semiconductor imaging chip camera.

[0133] Light source: In order to illuminate the inside of the digestive tract, the wireless capsule endoscope has a built-in light source. This light source is an LED lamp or other low-power, high-brightness light source. The design of the light source needs to take into account the uniformity and brightness of the lighting to ensure that the camera can capture a clear image. Specifically, 6 white light-emitting diode lighting sources are used.

[0134] Image sensor: The image sensor is responsible for converting the light signal captured by the camera into an electrical signal and further into digital image data. This sensor is a highly sensitive CMOS or CCD sensor that can capture high-quality images.

[0135] Wireless communication module: The wireless communication module is a key component for the wireless capsule endoscope to communicate with external devices (such as data receiving devices). It includes components such as radio frequency transmitters and antennas, and can wirelessly transmit image data outside the body.

[0136] Power system: The power system of the wireless capsule endoscope consists of batteries, which provide power to components such as the camera, light source, image sensor, and wireless communication module. The battery needs to have sufficient capacity and stability to ensure the working time of the capsule in the digestive tract. Specifically, the power source is 2 silver oxide batteries.

[0137] Control unit: Although wireless capsule endoscopes do not require a complex control unit, some advanced models contain a microprocessor or other control elements to control the camera's shooting parameters, the brightness of the light source, the rate of wireless communication, etc.

[0138] Other auxiliary components: The wireless capsule endoscope also contains other auxiliary components, such as temperature sensors, pressure sensors, etc., which are used to monitor the environmental parameters inside the digestive tract. These components provide additional diagnostic information to help doctors understand the patient's digestive tract condition more comprehensively.

[0139] The wireless capsule endoscope is a highly integrated medical device that realizes non-invasive examination of the human digestive tract through the coordinated work of components such as cameras, light sources, image sensors, and wireless communication modules. The data acquisition layer of the intelligent analysis system based on digestive tract inflammation collects data related to digestive tract inflammation through multiple channels and methods, and ensures the quality and accuracy of the data through data quality control measures. These data will provide a solid foundation for subsequent data processing and analysis.

[0140] 2. Data processing layer,

[0141] The data processing layer is one of the core components of the system. It is responsible for cleaning, integrating, and standardizing the raw data collected by the data acquisition layer so that the subsequent intelligent analysis layer can perform further analysis and mining. It includes the following functions:

[0142] Data cleaning: During the data collection process, the original data needs to be cleaned to remove duplicate, erroneous or irrelevant information to ensure the quality of the data.

[0143] Data integration: Integrate data from different systems to form a structured data set.

[0144] Feature extraction: Extract key features related to gastrointestinal inflammation from the integrated data, such as inflammatory marker levels, imaging features, etc.

[0145] Data standardization: Standardize the data to facilitate subsequent intelligent analysis.

[0146] Among them, data cleaning was performed using a non-information frame system for automatic detection of enteroscopy for digestive tract complications;

[0147] An effective computer-assisted system for automatically detecting and estimating disease features present in enteroscopy for digestive tract complications promotes standardized disease assessment. An important obstacle to automatically analyzing enteroscopy videos for digestive tract complications is the large number of non-informative frames caused by motion blur, blurred vision caused by debris, changes in light intensity and image. Those non-informative frames interrupt disease severity estimation by providing non-informative or conflicting information. An effective classification method is crucial to discard non-informative frames while retaining informative frames. The automatic detection system for enteroscopy for digestive tract complications has the following functions:

[0148] 2.1. Collect videos

[0149] Images of the digestive tract of patients with inflammatory bowel disease were collected. Videos were recorded at 30 frames per second at 1920×1080 resolution. Video frames sampled at a rate of one frame per second were manually annotated by a board-certified gastroenterologist. Specifically, the following four types of frames were annotated as non-informative:

[0150] A motion blurred frame is sufficient to blur the colon vasculature.

[0151] BFrames captured when the camera view is obscured by excessive liquid or solid debris.

[0152] C Frame captured when the camera was too close to the colon wall.

[0153] D The picture is overexposed due to a failure of automatic lighting control.

[0154] The frame distribution in each digestive tract inflammatory complication enteroscopic examination video was used to identify and remove the above four types of non-informative frames.

[0155] 2.2. Preprocessing

[0156] Frames from different digestive tract inflammatory complication enteroscopy examinations were preprocessed to ensure consistency. First, each frame was binarized, the largest component in the frame was identified by the content captured by the camera, and the smallest rectangular area containing the largest component was used to crop the original frame. After that, zero padding was added to fill the rectangular area into a square area, and then the size of the square area was resized to 256×256 pixels. The region of interest for feature extraction was an octagon.

[0157] 2.3. Bottleneck feature acquisition

[0158] Bottleneck features are used for non-informative frame classification. More than 10,000 annotated frames were used for model training. To avoid bias from individual patient characteristics, a pre-trained model was used and fine-tuned using annotated training data. The activation map from the last average pooling layer was considered as the bottleneck feature. The number of bottleneck features was 2048.

[0159] Feature extraction

[0160] The hue, saturation, and value channels in the HSV color space correspond to color, grayscale, and brightness, respectively. The hue channel is visualized by a color map. Unlike the RGB color space, the HSV color space separates color from intensity. Extracting features from frames in the HSV color space provides additional information. While many pre-trained models are available for generating features in the RGB color space, large-scale datasets suitable for HSV feature extraction are limited, where reflection, shadow, and brightness in an image should be relevant for classification tasks.

[0161] In order to overcome the shortcomings of large-scale datasets, the frames are converted from RGB to HSV, and a feature extraction algorithm is proposed. Features are extracted to characterize the frames in the digestive tract inflammatory colonoscopy video. The details of each feature extraction are as follows:

[0162] 2.4.1 Specular reflection mask establishment: Due to specular reflections from the surface of the colon, many frames contain many bright areas, which will significantly increase the number of edges and contrast of the image. To eliminate the effect of these reflections, a reflection mask is established for each frame using threshold and dilation operations on the saturated channel. The fraction of reflective areas in the region of interest is calculated as a feature. The reflection mask is used to correct features in other categories.

[0163] 2.4.2. Intensity Statistics: After removing the pixels inside the reflective mask and outside the region of interest, the mean, variance, skewness and kurtosis of the pixel intensities in the hue, saturation and value channels are calculated respectively. The mean gives the average level, while the variance calculates the heterogeneity. In addition, the skewness and kurtosis measure the asymmetry and tail of the distribution respectively. The intensity statistics characterize the distribution of color components, grayscale and brightness in the region of interest.

[0164] 2.4.3 Edge feature extraction: First, contrast-limited adaptive histogram equalization is used to enhance the contrast in the value channel, where the contrast transformation function is calculated on the local area. Secondly, the Canny edge detector is used to generate an initial edge mask. To eliminate the effect of high reflection, the edges within the reflection mask are removed. In addition, each edge component is analyzed, and edges with an eccentricity less than 0.9 are removed. The percentage of edges in the region of interest is calculated as an edge feature. The Hough transform is used to find lines in the edge. Considering that a typical information frame contains a clear view of the colon lumen and colon folds, the number of linear edges is also used as an edge feature. The fold contours of the colon are successfully detected. Edge detection using the Canny method will also find sharp intensity gradients in the colon wall (e.g., blood vessels in the colon wall), but line detection responses tend to gather at fold contours and significant disease features.

[0165] 2.4.4 Gray-level co-occurrence matrix calculation: In order to eliminate the influence of the boundary of the region of interest, inward interpolation is used to fill the area outside the region of interest. Similarly, in order to eliminate the influence of reflection, the area falling within the reflection mask is also filled. The gray-level co-occurrence matrix is ​​calculated for each channel separately. The pixel intensity is quantized into 32 levels. The second-order statistics of the gray-level co-occurrence matrix include contrast F con Energy F ener and uniformity F hom , calculated as follows:

[0166]

[0167] v(i,j) represents the gray level co-occurrence matrix value at the position (i,j).

[0168] 2.4.5 Blur calculation: The degree of blur is an important indicator for the classification of uninformative images. To measure blur, rectangular regions (sub-regions) obtained from inward interpolation are used. A Gaussian filter is used to generate an absolute difference map between the original and blurred sub-regions. The mean and standard deviation of the differences are calculated as blur metrics. This is based on the assumption that the Gaussian filter has a small effect on blurred images. In addition, the reference-free image quality measure of blurred images in the frequency domain proposed in is calculated. Since blurred images are also caused by optical defocus, a set of focus metrics are calculated. By convolving the discrete Laplacian mask with the focus measurement of the input image, the measurement value F LAPE Written as:

[0169]

[0170] Where I represents the input image, H and W represent the height and width of the input image respectively.

[0171] By dividing the image into multiple square blocks and calculating the energy ratio F of the AC and DC coefficients of each block in the discrete cosine transform domain DCTR , used to measure focus:

[0172]

[0173] Where S is the block size. u,v is the square of the AC coefficient in block (u, v), EDC u,v is the square of the DC coefficient in block (u, v). Preferably, S=15 is used.

[0174] 2.5. Classifier Training

[0175] A combination of handcrafted features and bottleneck features was used to train the random forest model. Considering the large number of features after feature fusion, the random forest model with automatic feature selection was chosen.

[0176] In the digestive tract inflammatory colonoscopy video, image clarity is affected by camera motion. In addition, image brightness and clarity are affected by variable characteristics of the internal colon environment, including the amount of water or debris, surface texture, and the distance between the camera and the colon wall. Therefore, image analysis in the HSV color model (separating color components from intensity) provides additional information. A set of hand-crafted features were extracted in the HSV color space, including measures of edges, reflections, blur, and focus. The hand-crafted features in the HSV color space and the RGB color space effectively improved the classification performance.

[0177] The proposed automatic detection of non-informative frames for enteroscopic examinations with GI complications is essential for further analysis of enteroscopic examinations with GI complications. Accurate detection and removal of non-informative frames can effectively improve the accuracy of disease severity estimation and reduce computational costs.

[0178] Data integration integrates data from different data sources and different formats to form a unified data set. Convert data in different formats (such as text, images, videos, etc.) into a unified format for subsequent processing. Use data integration technology (such as ETL tools, data warehouses, etc.) to merge data from different data sources and build a unified data view. Associate different data tables or data sets through key fields (such as patient ID, examination time, etc.) to achieve data interconnection.

[0179] Data standardization ensures data consistency and comparability, and provides a standardized data foundation for subsequent intelligent analysis. Classified data is uniformly coded, such as disease names and drug names, using internationally accepted coding standards (such as ICD, ATC, etc.). Numerical data is uniformly processed, such as length units converted to meters, weight units converted to kilograms, etc. For data that requires numerical calculations (such as age, laboratory indicators, etc.), normalization or standardization is performed to eliminate the impact of dimensional differences on analysis results.

[0180] In summary, the data processing layer of the intelligent analysis system based on gastrointestinal inflammatory complications provides high-quality and standardized input data for the intelligent analysis layer through steps such as data cleaning, data integration, and data standardization. This is an important guarantee for ensuring the accuracy and reliability of the system analysis results.

[0181] 3. Intelligent analysis layer

[0182] The intelligent analysis layer is the core of the entire system. It uses artificial intelligence technologies such as machine learning and deep learning to conduct in-depth analysis and mining of the data that has been cleaned, integrated, and standardized by the data processing layer to identify potential causes of gastrointestinal inflammation. It uses a curvelet-based wireless capsule endoscope image ulcer detection model, which is a multi-resolution analysis tool. Its performance is better than the wavelet analysis model in the two-dimensional domain. The selected color space is YCbCr. The image is first decomposed into multiple sub-bands of different scales and directions. Then, the porosity index of the selected sub-band is calculated and a feature vector is formed. The curvelet-based wireless capsule endoscope image ulcer detection model has the following functions

[0183] 3.1. Curvelet transform

[0184] The curvelet transform overcomes the weakness of wavelets and effectively represents two-dimensional singular points in images, such as lines and curves, which often appear in medical images. -j , direction θ l , position (j, l, k) and current time t, the moment when the image is at position (j, l, k) By translating and rotating the original curve The definitions are as follows:

[0185]

[0186] in, is the rotation θ l radians. The variable θ l is the rotation angle An equally spaced sequence of l ≤2π; is the sequence parameter of the translation. Through its Fourier transform is defined and represented in polar coordinates \((r,\theta)\) in the Fourier domain as:

[0187]

[0188] The support of is a polar parabolic wedge defined by the supports of \(W()\) and \(V()\) (radial and angular windows) respectively, applying a scale-dependent window width in each direction; \(W()\) and \(V()\) satisfy the partition of unity property. In continuous frequency \(v\), the curvelet coefficient \(\alpha\) of the two-dimensional function \(f\) j,l,k is defined as the inner product:

[0189]

[0190] There are two digital implementations of the discrete curvelet transform: by implementing curvelets non-uniformly, both running in an \(n\times n\) Cartesian array floating point and being invertible as well, with a fast inverse algorithm of approximately the same complexity.

[0191] 3.2. Differential porosity analysis

[0192] Porosity is a fractal property corresponding to the fractal dimension, used to measure how a fractal fills space, thus distinguishing textures and natural surfaces with the same fractal dimension. More specifically, porosity evaluates the distribution of gap sizes along a data set. A set of gaps of different sizes is considered heterogeneous, characterized by high porosity, while a homogeneous group with uniform gap sizes exhibits low porosity. It should be emphasized that a homogeneous set at a larger scale is very heterogeneous when examined at a smaller scale, and vice versa. From this perspective, porosity is considered a scale-dependent tool to measure the heterogeneity of textures.

[0193] The porosity is calculated using the sliding box algorithm, applicable to binary data sets. However, most real-life image analysis applications require extracting texture information from images, which cannot be binaryized. For this purpose, differential porosity applicable to gray-scale image analysis is introduced. Differential porosity is calculated by the differential box-counting algorithm, which uses a sliding box \(r\) of size \(r\times r\) pixels and a sliding window \(w\) of size \(w\times w\) pixels (\(r < w\)). The window scans the entire image, while the box scans the pixels bounded by the window to calculate the box quality at each position of the window. Depending on the pixel values contained in the box, more than one column of cubes of size \(r\times r\times r\) is required to cover the image intensity surface. Numbers 1, 2, etc. are assigned to the cubes from bottom to top, and the height difference \(n(i,j)\) of the columns is calculated. When the box slides across the window, the sum of all \(n(i,j)\) is the box quality of the window. Then, the differential porosity value \(L(r, w)\) is calculated through the same procedure as in the sliding box algorithm.

[0194] 3.3. Distinguishing tissues

[0195] A texture-based detection and classification scheme is developed to distinguish normal tissue from concurrently inflammatory tissue in wireless capsule endoscopy images. In the first step of the proposed method, the image is decomposed into its color channels. Similar work in the field of curvelet transform of wireless capsule endoscopy images has shown that the more perceptually uniform HSV and YCbCr spaces provide better accuracy than the commonly used hardware-oriented RGB space. YCbCr provides the best performance of all three color spaces. YCbCr is a family of color spaces used as part of the color image pipeline in video and digital photography systems. Y is the luminance component and Cb and Cr are the blue difference and red difference chrominance components. The Y, Cb, Cr channels are obtained from the R, G, B channels as shown below:

[0196] Y=16+(66R+129G+25B)

[0197] Cb=128+(-38R-74G+112·B)

[0198] Cr=128+(112R-94G-18B)

[0199] In the next step, the discrete curvelet transform is applied to each color channel by means of a winding method. Two-dimensional discontinuities such as straight lines and curves are very common phenomena in medical imaging. The multi-resolution and multi-directional properties of the discrete curvelet transform allow for the efficient detection of such singularities regardless of their orientation, size and location. The two main analysis parameters of the discrete curvelet transform, the number of scales and angles, are determined. Three or four analysis scales are usually used for related applications. The limiting factor in this case is that the size of the original image is quite small (256x256 pixels or less after segmentation). As the number of scales increases, the size of the subbands (in pixels) decreases, which eventually leads to information loss. In the method, after exhaustive trials, a four-scale decomposition was used. In addition, each scale contains a certain number of angles, which vary from scale to scale. Eight angles (must be a multiple of 4, with a minimum number of 8) were selected as the second scale. The use of 16 angles for the second scale was also tested. However, it was noted that information was repeated, which would lead to increased computational time and data redundancy. Specifically, when 16 angles are used for the second scale, the first 8 sub-images are very similar to the last 8 sub-images. The above selection of scales and angles results in 1, 8, 16, and 16 angles per scale, respectively, for a total of 41 discrete curvelet transformed sub-images.

[0200] The next step consists in calculating the differential porosity index for each subband. The differential porosity is calculated for a constant frame size r = 3 pixels and a window size w = 4frame_max, where frame_max is equal to the minimum size of each subimage. In this way, a differential porosity curve is obtained for each subimage, with as many coefficients as possible. Each differential porosity curve is normalized to the highest value to ensure the same reference level. However, the above calculations lead to feature vectors of high dimensionality, which can negatively affect the speed and accuracy of the classification process. For this purpose, a hyperbolic function is used to model the differential porosity curve Λ(w), defined as follows

[0201]

[0202] Parameters (a, b, c) are independent variables calculated as solutions to the least squares problem. These three parameters represent the differential porosity-based texture features of wireless capsule endoscopy images. , Six common statistical features, namely, mean, standard deviation, entropy, energy, skewness, and kurtosis, are also extracted from the differential porosity curves.

[0203] Three independent feature sets A, B and C were formed. Feature set A included only the (a, b, c) parameters of the modeling process. Set B was a combination of the (a, b, c) parameters of the hyperbolic function (global behavior of the expression curve) and the first three differential void fraction values ​​(global behavior of the expression curve). Set C included 6 statistical measurements obtained from the differential void fraction curve. The classification process was performed by a support vector machine with a radial basis function kernel and a unit scale factor. Ten-fold cross validation was used for training / testing of the classifier. In order to measure the classification performance, sensitivity and specificity indices were used together with the classification accuracy.

[0204] As mentioned above, for each set A, B or C, 41 different feature vectors are obtained, one for each of the 41 different angles (subbands) of the discrete curvelet transform. The purpose of this feature extraction method is to test the discrimination ability of each angle individually. For angle 4 of scale 4 in the Cr color channel, the highest classification accuracy achieved by set A is 77.57%. The Cr channel proved to be the most effective.

[0205] The Cr chroma component generally provides enhanced performance compared to Y and Cb, meaning that most ulcer information is more thoroughly expressed by the amount of blue-green or purple-red hues. This is because the reflected green light is closely related to blood volume and the intestinal wall is packed with blood vessels. In addition, not every angle performs equally well in distinguishing normal and ulcer images. More specifically, for all three sets, there are certain angles (with an offset of 1) that consistently show local peaks. For example, for the Cr channel, angles 1, 4, 8, 12 in scales 2 and 3 generally have local peaks or are very close to local peaks. Similar observations are made for color channels Y and Cb, namely, specific angles have consistently high classification accuracy.

[0206] By combining features from different angles, the classification accuracy of a single angle is improved. Several combinations of different numbers of angles have been tested. However, no improved classification results could be achieved using features extracted from more than 2 angles, and the testing was performed only on the Cr channel, as it provided the best results in the previous analysis. In this case, the highest performance is achieved by using the features of the angle that achieved the best classification rate with the features of one of the remaining angles (not necessarily by combining features from different angles to improve the classification accuracy of a single angle). Several combinations of different numbers of angles have been tested. However, no improved classification results can be achieved using features extracted from more than 2 angles, as one would expect, due to the curse of dimensionality. Testing was performed only on the Cr channel, as it provided the best results in the previous analysis. In this case, the highest performance is achieved by using the features of the angle that achieved the best classification rate with the features of one of the remaining angles. Set C produces the best results (86.54% accuracy) by using features from (angle 4 / scale 3) and (angle 1 / scale 1). The combination of (angle 5 / scale 3) with (angle 1 / scale 4) proved to be the most effective for set B, scoring 80.39% accuracy. In contrast, set A is the least efficient, achieving only 78.74% accuracy with features from (angle 4 / scale 4) and (angle 2 / scale 2).

[0207] To further demonstrate the performance of the discrete curvelet transform-differential porosity scheme, the above results are compared with those obtained after applying the proposed method, where the popular curvelet-based LBP analysis on the same dataset showed 82.2% acc, 87.1% sensitivity and 77% spec, which are 4.34% and 7.51% lower acc and spec, respectively. These results indicate that the extraction of differential porosity-based texture features of discrete curvelet transform subbands is more effective in ulcer detection than similar methods.

[0208] 4. Decision support layer

[0209] The decision support layer of the intelligent analysis system based on gastrointestinal comorbid inflammation is a key component of the system. It is mainly responsible for converting the analysis results generated by the intelligent analysis layer into suggestions and information that are instructive for clinical decision-making. Using the results output by the intelligent analysis layer, combined with clinical guidelines, expert experience and patient specific conditions, provide doctors with personalized treatment plan evaluation and suggestions. Perform risk assessment on the patient's condition, predict complications and disease progression trends. The decision support layer predicts the inflammatory activity of ulcerative colitis based on deep learning and endoscopic examination, and establishes a new deep learning-based evaluation system to evaluate wireless capsule images, accurately describing the severity and distribution of inflammatory activity through full-length intestinal wireless capsule videos. Includes the following functions:

[0210] 4.1 Image and video data annotation,

[0211] Image annotation was performed for images and videos from patients who underwent full laparoscopy and obtained wireless capsule images of all 5 intestinal segments, including clear white-light endoscopic images without image-enhanced endoscopy, feces, blur or halo. All images were evaluated by 4 wireless capsule physicians with 30, 11, 4 and 6 years of experience, respectively.

[0212] 4.2 Artificial Intelligence Algorithms,

[0213] The AI ​​architecture is divided into the following modules: image classification framework, video processing pipeline, and weighted evaluation system. Due to its reliability and effectiveness, the pre-trained convolutional neural network framework was selected as the main one. It is a feature extractor, the first step of the training protocol, and then a probability score ranging from 0 to 1 is created for each image by the trained CNN. A dedicated CNN model is designed for the video processing pipeline, which contains a visual clarity module and an image similarity determination module. This CNN model is used to pre-process highly complex data and obtain image sequences as input to the evaluation module. Therefore, the visual clarity model first detects all frames in the visual clarity stage to remove low-definition images. The similarity detection process judges the similarity of each frame with 4 adjacent images, and then removes images that are too similar to the previous and next frames to prevent these images from being repeatedly evaluated.

[0214] 4.3 Evaluation model development

[0215] The evaluation system consists of two parts: baseline evaluation model and weighted evaluation model.

[0216] 4.3.1 Baseline Evaluation Model

[0217] The baseline assessment model was used to determine a quick sketch of the full-length intestinal tract, which was expressed as follows:

[0218]

[0219] Among them, M i (0-3) is an approximate clinician estimate reflecting the severity of inflammation associated with a particular bowel segment. Length length and N are the length of the inflamed portion and the number of inflamed bowel segments (0-5). The values ​​obtained using this equation are determined subjectively by each endoscopist based on their experience and have considerable variability.

[0220] 4.3.2 Weighted Evaluation Model

[0221] A weighted evaluation model was established and applied to the image sequence. Based on clinical experience, the intestine was divided into a fixed number of regions, including 20 images in the cecum region, 20 images in the transverse colon region, 20 images in the descending colon region, 15 images in the sigmoid colon region, and 10 images in the rectum region. Each region contains a certain number of intestinal images, and the system calculates the area score Arsco according to the following formula:

[0222] Arsco=(Num mild PPV mild +2Num erate PPV erate +3Num severe PPV severe )

[0223] Among them, avg is the average value of the evaluation prediction of all endoscopic images for the image sequence, Numx represents the proportion of the evaluation results in the intestinal area, and PPVx (positive predictive value) is the statistical analysis value predicted by the AI ​​algorithm of the Mayo classification task, which measures the ratio of true positive cases to classified positive cases. Clinicians pay attention to highly severe inflammation during clinical diagnosis. In the calculation formula of area score, mild, erate and severe represent mild, moderate and severe, respectively.

[0224] After determining the score of each region of the internal segment, the segmental Segsco score is generated by the following linear combination:

[0225]

[0226] Among them, ratio and L represent the percentage of each inflammation severity and the number of image sequences of the intestinal segment, respectively.

[0227] 4.4 Statistical analysis

[0228] The primary outcome measure was defined as the accuracy of the AI ​​algorithm's predictions. The harmonic mean of the positive predictive value and the negative predictive value was chosen as the basis for evaluating the model's performance. The secondary outcome measures were the outputs of the evaluation system, including the results of the area assessment and the segment assessment. The distribution of values ​​for the area score was discussed. In addition, to verify the validity of the segment score, the validation part focused on the correlation between the weighted evaluation system and the observer's baseline assessment model. The agreement was evaluated by confusion matrix and other metrics, including positive predictive value, negative predictive value, sensitivity, and specificity.

[0229] 5. User Interaction Layer

[0230] Provide an intuitive and easy-to-use user interface to facilitate doctors and other medical personnel to interact with the system. Display analysis results and diagnostic suggestions in the form of charts, reports, etc. to help doctors quickly understand the condition and develop treatment plans. Support doctors to enter patient information, query historical records, modify diagnostic suggestions, and other operations to improve work efficiency. The user interaction layer of the intelligent analysis system based on gastrointestinal complication inflammation is a key part of information exchange and interaction between the system and users. It is mainly responsible for receiving user input, displaying the system's analysis results, and providing a user-friendly operation interface. The following is the detailed composition of the user interaction layer:

[0231] 5.1 User Interface Design

[0232] Intuitive: Use a simple and clear interface design to ensure that users can easily understand and operate it.

[0233] Responsiveness: Ensure that the system interface can quickly respond to user operations and improve user experience.

[0234] Customizability: Provide personalized setting options, allowing users to adjust the interface layout, font size, etc. according to their needs.

[0235] 5.2 Input Module

[0236] Data input: Supports multiple data input methods, such as text input, file upload (such as medical records, imaging materials, etc.), voice input, etc. Provides clear data input guidance to ensure that users can correctly and completely enter the required information.

[0237] Patient Information Management: Allows users to create, edit and query basic patient information such as name, age, gender, medical history, etc.

[0238] 5.3 Result display and feedback

[0239] Analysis result display: The results generated by the intelligent analysis layer are displayed to users in an intuitive manner, such as charts, reports, etc.

[0240] Detailed explanation: Provide detailed explanation and description of the analysis results to help users understand the cause, pathological process, treatment plan, etc.

[0241] Interactive feedback: Allow users to provide feedback on analysis results, such as marking errors, raising questions, etc., so that the system can be continuously optimized.

[0242] 5.4 Auxiliary tools and resources

[0243] Clinical Guidelines and Knowledge Base: Provide clinical guidelines, medical knowledge base and other resources related to gastrointestinal inflammation for users to refer to and learn from.

[0244] Case database: displays typical cases and successful cases to help users understand the treatment methods and effects of different diseases.

[0245] Online help and customer service: Provide online help documents, FAQ, customer service support, etc. to solve problems encountered by users during use.

[0246] 5.5 Security and Privacy Protection

[0247] Data encryption: Encrypt the data input by the user and the information during transmission to ensure data security.

[0248] Privacy protection: Strictly abide by relevant laws and regulations to protect users' personal privacy information from being leaked.

[0249] In summary, the user interaction layer provides users with a convenient, efficient and secure interaction platform through the composition of user interface design, input module, result display and feedback, auxiliary tools and resources, and security and privacy protection. The design and optimization of this layer is of great significance to improving user experience and overall system performance.

[0250] The systems, devices, modules or units described in the above embodiments may be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, the computer may be, for example, a personal computer, a laptop computer, a tablet computer, a wearable device, or a combination of any of these devices.

[0251] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0252] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0253] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

Claims

1. An intelligent analysis system based on digestive tract complications and inflammation, including five main parts: data collection layer, data processing layer, intelligent analysis layer, decision support layer and user interaction layer, among which, The data collection layer is responsible for collecting data from various channels, including hospital information systems, laboratory information systems, and image acquisition systems; the image acquisition system includes wireless capsule endoscopes, ultrasound wireless capsules, CT, and MRI imaging equipment, which are used to collect images of the digestive tract; The data processing layer is responsible for cleaning, integrating, and standardizing the raw data collected by the data collection layer so that the subsequent intelligent analysis layer can perform further analysis and mining; Data cleaning: During the data collection process, the original data needs to be cleaned to remove duplicate, erroneous or irrelevant information to ensure the quality of the data; Data integration: Integrate data from different systems to form a structured data set; Feature extraction: extract key features related to gastrointestinal inflammation from the integrated data; Data standardization: standardize the data for subsequent intelligent analysis; Among them, data cleaning was performed using a non-information frame system for automatic detection of enteroscopy for digestive tract complications; The intelligent analysis layer uses machine learning and deep learning artificial intelligence technologies to conduct in-depth analysis and mining of the data that has been cleaned, integrated, and standardized by the data processing layer to identify potential causes of gastrointestinal inflammation; it uses a curve-based wireless capsule endoscope image ulcer detection model; The decision support layer, based on the intelligent analysis system of digestive tract complications and inflammation, is a key component of the system. It is mainly responsible for converting the analysis results generated by the intelligent analysis layer into suggestions and information that are instructive for clinical decision-making; The user interaction layer provides an intuitive and easy-to-use user interface to facilitate doctors and other medical personnel to interact with the system. It displays analysis results and diagnostic suggestions in the form of charts and reports to help doctors quickly understand the condition and develop treatment plans. It supports doctors to input patient information, query historical records, and modify diagnostic suggestions to improve work efficiency. The user interaction layer of the intelligent analysis system based on gastrointestinal inflammatory complications is the key part for information exchange and interaction between the system and the user; it is mainly responsible for receiving user input, displaying the system's analysis results, and providing a user-friendly operation interface.

2. The intelligent analysis system based on digestive tract inflammation according to claim 1, characterized in that: The specific components of wireless capsule endoscope are as follows: Shell: The shell of the wireless capsule endoscope is made of materials that meet biocompatibility standards, including medical-grade plastics or metal alloys; this shell not only protects the internal components from the external environment, but also ensures the safe passage of the capsule in the digestive tract; the shell design takes into account the patient's swallowing comfort and the stability of the capsule in the digestive tract; Camera: The camera is one of the core components of the wireless capsule endoscope, responsible for capturing images of the inside of the digestive tract; It is a miniature camera with high resolution and wide-angle field of view, which can clearly capture the fine structure of the digestive tract mucosa; the camera is combined with the optical system to ensure the quality and clarity of the image; Light source: To illuminate the inside of the digestive tract, the wireless capsule endoscope has a built-in light source; This light source is an LED lamp or other low-power, high-brightness light source; the design of the light source needs to take into account the uniformity and brightness of the lighting to ensure that the camera can capture a clear image; Image sensor: The image sensor is responsible for converting the light signal captured by the camera into an electrical signal and further into digital image data; this sensor is a highly sensitive CMOS or CCD sensor that can capture high-quality images; Wireless communication module: The wireless communication module is a key component for the wireless capsule endoscope to communicate with external devices; it includes a radio frequency transmitter and an antenna element, which can wirelessly transmit image data outside the body; Power system: The power system of the wireless capsule endoscope consists of batteries, which provide power for the camera, light source, image sensor and wireless communication module components; The battery needs to have sufficient capacity and stability to ensure the capsule's operating time in the digestive tract; Control unit: Although wireless capsule endoscopes do not require a complex control unit, some advanced models contain a microprocessor or other control elements to control the camera's shooting parameters, the brightness of the light source, and the rate of wireless communication; Other auxiliary components: The wireless capsule endoscope also contains other auxiliary components, including temperature sensors and pressure sensors, which are used to monitor the environmental parameters inside the digestive tract; these components provide additional diagnostic information to help doctors more fully understand the patient's digestive tract condition.

3. The intelligent analysis system based on digestive tract inflammation according to claim 1, characterized in that: The non-information frame system for automatic detection of digestive tract complications and enteroscopic examinations has the following functions: 2.1 Collect Video Images of the digestive tract of patients with inflammatory bowel disease were collected; videos were recorded at 30 frames per second at 1920 × 1080 resolution; video frames sampled at a rate of one frame per second were manually annotated by a board-certified gastroenterologist; specifically, the following four types of frames were annotated as non-informative: A frame with motion blur sufficient to blur the colon vasculature; B Frames captured when the camera view is obscured by excessive liquid or solid debris; C Frame captured when the camera is too close to the colon wall; D The picture is overexposed due to failure of automatic lighting control; The distribution of frames in each digestive tract inflammatory complication enteroscopy video used was used to identify and remove the above four types of non-informative frames; 2.2 Preprocessing Frames from different enteroscopic examinations of digestive tract complications were preprocessed to ensure consistency; first, each frame was binarized, and the largest component in the frame was identified by the content captured by the camera. The smallest rectangular area containing the largest component was used to crop the original frame; after that, zero padding was added to fill the rectangular area into a square area, and then the size of the square area was resized to 256×256 pixels. The region of interest for feature extraction was an octagon; 2.3 Bottleneck Feature Acquisition Bottleneck features were used for non-informative frame classification; more than 10,000 annotated frames were used for model training. To avoid bias in individual patient characteristics, a pre-trained model was used and fine-tuned using annotated training data. The activation map from the last average pooling layer was considered as the bottleneck feature. The number of bottleneck features was 2048. 2.4 Feature Extraction The hue, saturation, and value channels in the HSV color space correspond to color, grayscale, and brightness, respectively; the hue channel is visualized through a color map; unlike the RGB color space, the HSV color space separates color from intensity; Extracting features from frames in the HSV color space provides additional information; while many pre-trained models can be used to generate features in the RGB color space, large-scale datasets suitable for HSV feature extraction are limited, where reflections, shadows, and brightness in the image should be relevant to the classification task; 2.5 Classifier Training A combination of handcrafted features and bottleneck features was used to train a random forest model; considering the large number of features after feature fusion, a random forest model with automatic feature selection was selected; In the digestive tract inflammatory colonoscopy videos, image clarity is affected by camera motion; in addition, image brightness and clarity are also affected by variable characteristics of the internal colon environment, including the amount of water or debris, surface texture, and the distance between the camera and the colon wall; therefore, image analysis in the HSV color model provides additional information; a set of hand-crafted features are extracted in the HSV color space, including measures of edges, reflections, blur, and focus, and the hand-crafted features in the HSV color space and the RGB color space effectively improve the classification performance.

4. The intelligent analysis system based on digestive tract inflammation according to claim 3, characterized in that: Extract features to characterize the frames in the digestive tract inflammatory colonoscopy video; the details of each feature extraction are as follows: 2.4.1 Specular reflection mask establishment: Due to specular reflections from the surface of the colon, many frames contain many bright areas, which will significantly increase the number of edges and contrast of the image; To eliminate the effects of these reflections, a reflection mask is established for each frame using threshold and dilation operations on the saturated channel; The score of the reflective area in the region of interest is calculated as a feature; The reflection mask is used to correct features in other categories; 2.4.

2. Intensity Statistics: After removing pixels within the reflective mask and pixels outside the region of interest, the mean, variance, skewness and kurtosis of the pixel intensities in the hue, saturation and value channels are calculated respectively; the mean gives the average level, while the variance calculates the heterogeneity; in addition, the skewness and kurtosis measure the asymmetry and tail of the distribution respectively; the intensity statistics characterize the distribution of color components, grayscale and brightness in the region of interest; 2.4.3 Edge feature extraction: First, contrast-limited adaptive histogram equalization is used to enhance the contrast in the value channel, where the contrast transformation function is calculated on the local area; second, the Canny edge detector is used to generate an initial edge mask; to eliminate the influence of high reflection, the edges within the reflection mask are removed; in addition, each edge component is analyzed, and the edges with eccentricity less than 0.9 are removed; the percentage of edges in the region of interest is calculated as an edge feature; The Hough transform was used to find lines in the edges; considering that a typical information frame contains a clear view of the colon lumen and colon folds, the number of linear edges was also used as an edge feature; The fold contours of the colon are successfully detected; edge detection using the Canny method will also find sharp intensity gradients in the colon wall, but line detection responses tend to cluster at fold contours and prominent disease features; 2.4.4 Gray-level co-occurrence matrix calculation: In order to eliminate the influence of the boundary of the region of interest, inward interpolation is used to fill the area outside the region of interest; The areas falling within the reflectance mask were also filled, and the gray-level co-occurrence matrix was calculated for each channel separately; the pixel intensity was quantized into 32 levels; The second-order statistics of the gray-level co-occurrence matrix include contrast, energy, and uniformity calculations; 2.4.5 Fuzziness calculation: The degree of blur is an important indicator for the classification of uninformative images; to measure blur, a rectangular region obtained from inward interpolation is used; a Gaussian filter is used to generate an absolute difference map between the original and blurred sub-regions; the mean and standard deviation of the differences are calculated as blur metrics; This is based on the assumption that Gaussian filters have less impact on blurred images; In addition, the no-reference image quality measures for blurred images in the frequency domain proposed in are computed; since blurred images are also caused by optical defocus, a set of focus metrics are computed; focus measures are obtained by convolving the discrete Laplacian mask with the input image.

5. The intelligent analysis system based on digestive tract inflammation according to claim 1, characterized in that: The curvelet-based wireless capsule endoscope image ulcer detection model has the following functions: 3.1 Curvelet Transformation,The second generation curvelet is achieved by translating and rotating the original curvelet; 3.2 Differential porosity analysis,The sliding box algorithm is used to calculate the porosity, which is applicable to binary data sets, and the differential porosity suitable for grayscale image analysis is introduced; 3.3 Differentiating tissues,By developing a texture-based detection and classification scheme to,distinguish normal tissues and concurrent inflammatory tissues in,wireless capsule endoscopy images, the images are decomposed into color channels,,Y,is the brightness component and Cb and Cr,are the blue difference and red difference chrominance components;,Y, Cb, and Cr channels are obtained from R, G, and B channels.

6. The intelligent analysis system based on digestive tract inflammation according to claim 1, characterized in that: A new deep learning-based evaluation system was established to evaluate wireless capsule images, accurately describing the severity and distribution of inflammatory activity through full-length intestinal wireless capsule videos; including the following functions: The decision support layer includes the following functions: 4.1 Image and Video Data Annotation Image annotation was performed on images and videos from patients who underwent full laparoscopy and obtained wireless capsule images of all 5 intestinal segments, including clear white-light endoscopic images without image-enhanced endoscopy, stool, blur, or halos; all images were evaluated by 4 wireless capsule physicians with 30, 11, 4, and 6 years of experience, respectively; 4.2 Artificial Intelligence Algorithms The AI ​​structure is divided into the following modules: image classification framework, video processing pipeline and weighted evaluation system; Due to its reliability and effectiveness, the pre-trained convolutional neural network framework was selected as the main one, which is a feature extractor, the first step of the training protocol, and then a probability score ranging from 0 to 1 is created for each image by the trained CNN; A dedicated CNN model is designed for the video processing pipeline, which contains a visual clarity module and an image similarity determination module; This CNN model is used to pre-process highly complex data and obtain image sequences as input to the evaluation module; Therefore, the visual clarity model first detects all frames in the visual clarity stage to remove low-definition images; The similarity detection process judges the similarity of each frame with 4 adjacent images, and then removes images that are too similar to the previous and next frames to prevent these images from being repeatedly evaluated; 4.3 Evaluation model development The evaluation system consists of two parts: baseline evaluation model and weighted evaluation model; 4.4 Statistical analysis The primary outcome measure was defined as the accuracy of the AI ​​algorithm's predictions; the harmonic mean of the positive and negative predictive values ​​was chosen as the basis for evaluating the model's performance; the secondary outcome measures were the outputs of the assessment system, including the results of the area and segmental assessments; the distribution of values ​​for the area scores was discussed; in addition, to verify the validity of the segmental scores, the validation part focused on the correlation between the weighted assessment system and the observer's baseline assessment model; Agreement was assessed using confusion matrices and other metrics, including positive predictive value, negative predictive value, sensitivity, and specificity.

7. The intelligent analysis system based on digestive tract inflammation according to claim 6, characterized in that: 4.3.1 Baseline assessment model, the baseline assessment model uses a method to determine the whole-length intestinal rapid sketch, reflecting the severity of inflammation in specific intestinal segments and the number of intestinal segments with inflammation; 4.3.2 Weighted Evaluation Model A weighted evaluation model was established and applied to image sequences. Based on clinical experience, the intestine was divided into a fixed number of regions, including 20 images in the cecum region, 20 images in the transverse colon region, 20 images in the descending colon region, 15 images in the sigmoid colon region, and 10 images in the rectum region. Each region contained a certain number of intestinal images. After determining the score of each region of the internal segment, the segmentation score was generated by the following linear combination.

8. The intelligent analysis system based on digestive tract inflammation according to claim 1, characterized in that: 5.1 User interface design, intuitive: Use a simple and clear interface design to ensure that users can easily understand and operate; Responsiveness: Ensure that the system interface can quickly respond to user operations and improve user experience; Customizability: Provide personalized setting options to allow users to adjust the interface layout and font size according to their needs; 5.2 Input module, data input: supports multiple data input methods, including text input, file upload, and voice input; provides clear data input guidance to ensure that users can correctly and completely enter the required information; Patient information management: allows users to create, edit and query basic information of patients, including name, age, gender, and medical history; 5.3 Result display and feedback, analysis result display: the results generated by the intelligent analysis layer are displayed to users in an intuitive way, including charts and reports; Detailed explanation: Provide detailed explanation and description of the analysis results to help users understand the cause, pathological process, and treatment plan; Interaction Feedback: Allow users to provide feedback on analysis results, including marking errors and raising questions, so that the system can be continuously optimized; 5.4 Auxiliary tools and resources, clinical guidelines and knowledge base: Provide clinical guidelines and medical knowledge base resources related to gastrointestinal inflammation for users to refer to and learn; Case database: displays typical cases and successful cases to help users understand the treatment methods and effects of different diseases; Online help and customer service: provides online help documents, FAQ, customer service support to solve problems encountered by users during use; 5.5 Security and privacy protection, data encryption: encrypt the data input by the user and the information during transmission to ensure data security; privacy protection: strictly abide by relevant laws and regulations to protect the user's personal privacy information from being leaked.

Images