Gastric mucosa image analysis method and system for Helicobacter pylori detection

Abstract

The present invention relates to the field of image processing technology, and in particular to a gastric mucosa image analysis method and system for Helicobacter pylori detection. The method comprises the following steps: acquiring gastric mucosa image data; performing image segmentation on the gastric mucosa image data to obtain gastric mucosa image segmentation data, and performing regional characteristic processing based on the gastric mucosa image segmentation data to obtain regional characteristic data; performing image enhancement on the gastric mucosa image data based on the regional characteristic data to obtain gastric mucosa image enhancement data; acquiring region of interest data, and simulating the gastric mucosa image enhancement data based on the region of interest data to obtain region of interest simulation data, so as to perform gastric mucosa image analysis operations. The present invention provides more accurate image analysis data by performing more detailed processing operations on gastric mucosa images.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to a gastric mucosa image analysis method and system for Helicobacter pylori detection. Background Art

[0002] Image analysis generally uses mathematical models combined with image processing techniques to analyze the underlying features and upper structures, thereby extracting information with a certain degree of intelligence. Image analysis focuses more on studying the content of the image, including but not limited to the use of various image processing techniques. It is more inclined to analyze, interpret and identify the content of the image. The main process of image analysis is: input, segmentation, recognition, and interpretation. Conventional gastric mucosal image analysis methods for Helicobacter pylori detection often require the features and rules to be defined in advance, which is difficult to adapt to complex and changing scenes. Summary of the invention

[0003] In order to solve the above technical problems, the present invention proposes a gastric mucosal image analysis method and system for Helicobacter pylori detection to solve at least one of the above technical problems.

[0004] The present application provides a gastric mucosal image analysis method for Helicobacter pylori detection, comprising the following steps:

[0005] Step S1: Acquire gastric mucosa image data;

[0006] Step S2: performing image segmentation on the gastric mucosa image data to obtain gastric mucosa image segmentation data, and performing regional characteristic processing on the gastric mucosa image segmentation data to obtain regional characteristic data;

[0007] Step S3: performing image enhancement on the gastric mucosa image data according to the regional characteristic data to obtain gastric mucosa image enhancement data;

[0008] Step S4: Acquire the region of interest data, and simulate the gastric mucosa image enhancement data according to the region of interest data to obtain the region of interest simulation data for performing gastric mucosa image analysis operations.

[0009] In the present invention, through image segmentation and regional characteristic processing, the method can more accurately identify different regions and characteristics in the gastric mucosa, thereby improving the accuracy of Helicobacter pylori detection and facilitating the timely detection and treatment of early Helicobacter pylori infection. Image enhancement is performed based on regional characteristic data, so that each region of interest can be visually enhanced, thereby improving the doctor's ability to interpret and analyze the image. Through image segmentation and regional characteristic processing, the method can more accurately identify the region of interest, thereby improving the accuracy of gastric mucosa analysis. Through image enhancement and simulation, the method can provide richer image information, including multi-channel color mapping and simulation data, and can provide rich multi-dimensional data to adapt to the gastric mucosal image data collected under different circumstances.

[0010] Preferably, step S1 specifically includes:

[0011] Step S11: collecting gastric mucosa images through a high-resolution sensor to obtain primary gastric mucosa image data;

[0012] Step S12: performing standardization processing on the primary gastric mucosa image data to obtain standard gastric mucosa image data;

[0013] Step S13: performing metadata annotation on the standard gastric mucosa image data to obtain gastric mucosa image data.

[0014] In the present invention, by collecting images with a high-resolution sensor, detailed and high-quality gastric mucosal image data can be obtained, providing a reliable basis for subsequent analysis. Standardization ensures the consistency of the image, helps to eliminate changes caused by different acquisition equipment or environmental conditions, and makes the data easier to compare and analyze with other data. Metadata annotation enriches the additional information of the image, which helps to better manage, retrieve and identify the image.

[0015] Preferably, step S13 is specifically:

[0016] Step S131: performing tissue recognition on the standard gastric mucosa image data to obtain tissue recognition data;

[0017] Step S132: generating a tissue distribution description for the standard gastric mucosa image data according to the tissue identification data to obtain tissue distribution description data;

[0018] Step S133: in the process of gastric mucosa image acquisition, image acquisition parameter data is acquired;

[0019] Step S134: metadata annotation is performed on the standard gastric mucosa image data according to the image acquisition parameter data and the tissue distribution description data to obtain gastric mucosa image data.

[0020] The present invention improves the efficiency and accuracy of image analysis through automated tissue recognition, and can more accurately locate and describe tissue structures in images. Generating tissue distribution description data helps users better understand the location and characteristics of different tissues in the image, thereby providing more information. Metadata annotation makes image data more informative, helps medical experts better understand and use images, and improves the efficiency of data management and utilization. Associating image acquisition parameters with tissue distribution descriptions helps users better understand the background information of the image, which is helpful for image analysis.

[0021] Preferably, step S2 specifically includes:

[0022] Step S21: performing multi-scale image segmentation on the gastric mucosa image data to obtain gastric mucosa image segmentation data;

[0023] Step S22: performing regional characteristic processing according to the gastric mucosa image segmentation data to obtain regional characteristic data.

[0024] The multi-scale image segmentation in the present invention helps to separate different tissues and structures in the image, providing a clear basis for further analysis. Regional characteristic processing provides users with more detailed information about different tissues or structures by extracting characteristic information of each region, which helps more accurate analysis and diagnosis. The present invention can deeply understand gastric mucosal images, thereby providing users with more information and promoting more accurate image analysis.

[0025] Preferably, step S21 is specifically:

[0026] Step S211: performing multi-scale filtering on the gastric mucosa image data to obtain multi-scale gastric mucosa image data;

[0027] Step S212: performing texture feature extraction and gradient feature extraction on the multi-scale gastric mucosa image data to obtain gastric mucosa image texture feature data and gastric mucosa image gradient feature data;

[0028] Step S213: performing scale selection according to the gastric mucosa image texture feature data and the gastric mucosa image gradient feature data to obtain scale selection data;

[0029] Step S214: performing edge segmentation on the multi-scale gastric mucosa image data according to the scale selection data to obtain gastric mucosa image segmentation data.

[0030] The present invention introduces scale selection data, and adaptively selects a suitable scale level according to the local texture and gradient features of the image. Traditional methods are difficult to adapt to multiple structures of different scales and complexities at the same time, resulting in under-segmentation or over-segmentation problems. The present invention avoids the scale selection problem in the traditional method and helps to better adapt to diverse structures. By extracting the texture features and gradient features of the image, the present invention can consider the image information more comprehensively, and improve the precision and accuracy of the segmentation. The scale selection process is automated, does not need to rely on manual parameter selection, reduces the interference of subjective factors, and improves the intelligence of the algorithm. By performing multi-scale filtering on the gastric mucosal image data to generate multi-scale gastric mucosal image data, it is helpful to capture the features and structures of different scales in the image, and a variety of information from microscopic to macroscopic can be effectively included in the analysis, which improves the comprehensiveness of image segmentation. Extracting the texture features and gradient features of the gastric mucosal image data makes the segmentation process more informative. The texture features can capture the details and texture information of the image, while the gradient features can help determine the edges and contours. The combination of the two improves the accuracy of segmentation. Using the texture features and gradient features of gastric mucosal images for scale selection makes the segmentation process more adaptive, helps overcome the limitation of relying on fixed scales in traditional methods, and improves the accuracy of segmentation. Edge segmentation of multi-scale gastric mucosal image data based on scale selection data helps to accurately locate the structure and regional boundaries in gastric mucosal images and provide more accurate segmentation results.

[0031] Preferably, step S22 is specifically:

[0032] Performing color feature extraction, texture feature extraction, shape feature extraction, and spatial relationship feature extraction on the gastric mucosa image segmentation data to obtain gastric mucosa image color feature data, gastric mucosa image texture feature data, gastric mucosa image shape feature data, and gastric mucosa image spatial relationship feature data;

[0033] Performing feature selection on the gastric mucosa image color feature data, the gastric mucosa image texture feature data, the gastric mucosa image shape feature data, and the gastric mucosa image space relationship feature data to obtain color feature selection data, texture feature selection data, shape feature selection data, and space relationship feature selection data;

[0034] The color feature selection data, the texture feature selection data, the shape feature selection data and the spatial relationship feature selection data are fused to obtain the regional feature data.

[0035] In the present invention, by extracting multidimensional characteristics such as color, texture, shape and spatial relationship from the gastric mucosal image segmentation data, various information of the image can be fully captured, including color distribution, texture characteristics, shape contours and spatial relationships between objects, so as to more comprehensively describe the characteristics of each segmented area. After extracting the multidimensional characteristics, screening and combining the data according to the characteristic selection helps to optimize the combination of characteristics, select the most relevant and useful characteristics, reduce redundant information and improve the accuracy of segmentation. By fusing the color, texture, shape and spatial relationship characteristic data after characteristic selection, the information between different characteristics can be correlated to provide more comprehensive regional characteristic data, which helps to more accurately describe each segmented area.

[0036] Preferably, the gastric mucosa image enhancement data includes first gastric mucosa image enhancement data and second gastric mucosa image enhancement data, and step S3 is specifically as follows:

[0037] Perform weighted reset reliability calculation based on regional characteristic data to obtain regional characteristic weighted reset reliability data;

[0038] When it is determined that the regional characteristic weighted reset confidence data is less than or equal to the preset weighted reset confidence threshold data, local noise reduction strategy data is generated according to the regional characteristic data, and the gastric mucosa image data is subjected to noise reduction processing according to the local noise reduction strategy data to obtain first gastric mucosa image enhancement data;

[0039] When it is determined that the regional characteristic weighted reset confidence data is greater than the preset weighted reset confidence threshold data, high-pass filtering is performed on the gastric mucosa image data according to the regional characteristic data to obtain gastric mucosa image high-pass filtered data;

[0040] Performing spatial filtering on the high-pass filtered data of the gastric mucosa image to obtain spatial filtered data of the gastric mucosa image;

[0041] Dynamic range compression is performed on the spatially filtered data of the gastric mucosa image to obtain second gastric mucosa image enhanced data.

[0042] In the present invention, when the regional characteristic weighted reset confidence data is less than or equal to the preset weighted reset confidence threshold data, a local noise reduction strategy is used to process the image, which helps to remove noise and unnecessary details and improve the quality and readability of the image. When the regional characteristic weighted reset confidence data is greater than the preset weighted reset confidence threshold data, a high-pass filter is used to process the image, and the high-pass filter helps to enhance the edges and details in the image, making the target structure more clearly visible. Spatial filtering and dynamic range compression are performed on the high-pass filtered image, which helps to further improve the quality and contrast of the image, making the image more visually attractive and highlighting important features. Two different gastric mucosal image enhancement data are generated, the first gastric mucosal image enhancement data is mainly used for local noise reduction, and the second gastric mucosal image enhancement data is mainly used for edge and detail enhancement. The dual enhancement strategy can better meet the needs of different application scenarios.

[0043] Preferably, step S4 is specifically:

[0044] Step S41: acquiring region of interest data, and segmenting the gastric mucosa image enhancement data according to the region of interest data to obtain gastric mucosa region of interest image data;

[0045] Step S42: performing multimodal color mapping on the gastric mucosa region of interest image data to obtain region of interest image color mapping data;

[0046] Step S43: performing contrast enhancement on the image color mapping data of the region of interest to obtain enhanced image data of the region of interest;

[0047] Step S44: sharpening the image enhancement data of the region of interest to obtain sharpened image data of the region of interest;

[0048] Step S45: mapping the region of interest image sharpening data to preset pseudo color region image data, thereby obtaining region of interest image pseudo color data;

[0049] Step S46: Perform three-dimensional model simulation on the pseudo-color data of the region of interest image to obtain simulated data of the region of interest for performing gastric mucosa image analysis.

[0050] The present invention obtains region of interest data, and segments the gastric mucosal image enhancement data based on these data, which can accurately extract specific areas of concern to doctors or analysts, help focus on important areas, save time and improve efficiency. The multimodal color mapping strategy can better highlight the various information in the region of interest, so that different information can be distinguished in visualization, which helps doctors understand the various features in the image more clearly. Contrast enhancement and sharpening of the region of interest image data helps to enhance the details and edges in the image, making the region of interest clearer and more prominent. Mapping the region of interest image sharpening data to pseudo-color region image data can provide more vivid and intuitive visualization results, which helps to more easily identify and analyze different regions. The three-dimensional model simulation performed in the present invention can provide region of interest simulation data with more depth and stereoscopic sense, which helps users to more comprehensively understand and analyze the structure and features in the image.

[0051] Preferably, step S42 is specifically:

[0052] Performing structural recognition on the gastric mucosa region of interest image data to obtain region of interest structure type data, wherein the region of interest structure type data includes region of interest blood vessel data, region of interest structure data, and region of interest metabolic data;

[0053] Perform green channel mapping on the gastric mucosa region of interest image data according to the region of interest blood vessel data to obtain first region of interest image color mapping data;

[0054] Performing red channel mapping on the gastric mucosa region of interest image data according to the region of interest structure data to obtain second region of interest image color mapping data;

[0055] Performing blue channel mapping on the gastric mucosa region of interest image data according to the region of interest metabolism data to obtain third region of interest image color mapping data;

[0056] The first region of interest image color mapping data, the second region of interest image color mapping data and the third region of interest image color mapping data are synthesized to obtain the region of interest image color mapping data.

[0057] In the present invention, by performing structural recognition on the image data of the region of interest, different types of structures, such as blood vessels, structures and metabolism, can be distinguished. Then, these types of structures are mapped to different color channels, such as green, red and blue channels. Multimodal color mapping helps users to more clearly distinguish and understand different structures and information in the image. By using different channels for mapping, specific types of structural information can be highlighted. For example, mapping blood vessels to the green channel makes them more eye-catching in the image. Mapping structural information to the red channel and metabolic information to the blue channel helps users to better distinguish and analyze the characteristics of different structures. Conventional methods usually use a single image processing technology, such as segmentation or enhancement, which does not make full use of the multi-channel information of the image, resulting in the omission of important structures and features, limiting the user's comprehensive understanding of the image. The present invention uses multi-channel color mapping to provide a variety of visual information to help users more easily identify the features of the region of interest. The data of each channel contains different information, and the comprehensive use can improve the accuracy of data analysis. By synthesizing the mapping data of different channels into an image, a more comprehensive visualization result can be provided, which helps users see the relationship between different structures in one image and more comprehensively understand the features of the region of interest.

[0058] Preferably, the present application also provides a gastric mucosa image analysis system for Helicobacter pylori detection, which is used to execute the gastric mucosa image analysis method for Helicobacter pylori detection as described above, and the gastric mucosa image analysis system for Helicobacter pylori detection comprises:

[0059] A gastric mucosa image data acquisition module, used for acquiring gastric mucosa image data;

[0060] A regional characteristic processing module is used to perform image segmentation on the gastric mucosa image data to obtain gastric mucosa image segmentation data, and perform regional characteristic processing based on the gastric mucosa image segmentation data to obtain regional characteristic data;

[0061] An image enhancement module, used for performing image enhancement on gastric mucosa image data according to regional characteristic data to obtain gastric mucosa image enhancement data;

[0062] The gastric mucosa image analysis module is used to obtain the data of the region of interest, and simulate the gastric mucosa image enhancement data according to the data of the region of interest to obtain the simulated data of the region of interest for performing gastric mucosa image analysis operations.

[0063] The beneficial effects of the present invention are: through image segmentation and regional characteristic processing, the method can more accurately identify the region of interest, thereby improving the accuracy of gastric mucosa analysis. Through image enhancement and simulation, the method can provide richer image information, including multi-channel color mapping and simulation data, so that users can have a more comprehensive understanding of the condition of the gastric mucosa, so as to better make accurate analysis. The image segmentation, regional characteristic processing and image enhancement steps in the method can be partially or completely automated, thereby reducing the workload of users engaged in related work, improving analysis efficiency, and is very useful for clinical applications that process a large number of gastric mucosal images. The simulation of the region of interest data can be customized according to the specific situation of each user, realizing personalized data analysis, thereby overcoming the differences between different users. By providing more accurate and rich image data, the method helps to make accurate data analysis early and avoids the delay caused by complex biochemical reaction processes. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] Other features, objects and advantages of the present application will become more apparent by reading the detailed description of non-limiting implementations made with reference to the following drawings:

[0065] Figure 1 A flowchart showing the steps of a gastric mucosa image analysis method for Helicobacter pylori detection according to an embodiment;

[0066] Figure 2 A flowchart showing the steps of a method for collecting gastric mucosa image data according to an embodiment is shown;

[0067] Figure 3 A flowchart showing the steps of a metadata annotation method according to an embodiment is shown;

[0068] Figure 4 A flowchart of the steps of a multi-scale image segmentation method according to an embodiment is shown. DETAILED DESCRIPTION

[0069] The following is a clear and complete description of the technical method of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by technicians in this field without creative work are within the scope of protection of the present invention.

[0070] In addition, the accompanying drawings are only schematic illustrations of the present invention and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. The functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor methods and / or microcontroller methods.

[0071] It should be understood that, although the terms "first", "second", etc. may be used herein to describe various units, these units should not be limited by these terms. These terms are used only to distinguish one unit from another unit. For example, without departing from the scope of the exemplary embodiments, the first unit may be referred to as the second unit, and similarly the second unit may be referred to as the first unit. The term "and / or" used herein includes any and all combinations of one or more of the listed associated items.

[0072] See also Figures 1 to 4 The present application provides a gastric mucosal image analysis method for Helicobacter pylori detection, comprising the following steps:

[0073] Step S1: Acquire gastric mucosa image data;

[0074] Specifically, a high-resolution endoscope or a gastric mucosal scanner or other equipment is used to collect gastric mucosal image data in the user's stomach, including different areas and perspectives of the internal stomach.

[0075] Step S2: performing image segmentation on the gastric mucosa image data to obtain gastric mucosa image segmentation data, and performing regional characteristic processing on the gastric mucosa image segmentation data to obtain regional characteristic data;

[0076] Specifically, image segmentation methods of various scales are used, such as edge detection, region growing, or deep learning-based methods, to obtain segmentation data of the gastric mucosa. Characteristics such as color, texture, shape, and spatial relationship are extracted for each segmented region. For example, the color histogram, texture statistics, perimeter, and area of ​​each region can be calculated.

[0077] Step S3: performing image enhancement on the gastric mucosa image data according to the regional characteristic data to obtain gastric mucosa image enhancement data;

[0078] Specifically, weighted confidence calculation: According to the regional characteristic data, the weighted confidence of each region is calculated based on various factors, such as color consistency, texture consistency, etc. Local noise reduction strategy: For regions with low weighted confidence, local noise reduction strategies such as median filtering or wavelet transform noise reduction can be applied to reduce noise. High-pass filtering: For regions with high weighted confidence, high-pass filtering can be applied to enhance edges and details. Dynamic range compression: Improve the contrast of the image by adjusting the grayscale level of pixels.

[0079] Step S4: Acquire the region of interest data, and simulate the gastric mucosa image enhancement data according to the region of interest data to obtain the region of interest simulation data for performing gastric mucosa image analysis operations.

[0080] Specifically, region of interest selection: Based on the output of the algorithm, select specific regions of interest that contain parts that need further analysis. Simulation algorithm: Based on the characteristics of the region of interest, a simulation algorithm, such as a mathematical model or simulation technology, is used to generate simulation data to represent the different characteristics of the region. 3D model simulation: For areas that require 3D representation, 3D modeling technology can be used to simulate the structure and characteristics of the region of interest.

[0081] In the present invention, through image segmentation and regional characteristic processing, the method can more accurately identify different regions and characteristics in the gastric mucosa, thereby improving the accuracy of Helicobacter pylori detection and facilitating the timely detection and treatment of early Helicobacter pylori infection. Image enhancement is performed based on regional characteristic data, so that each region of interest can be visually enhanced, thereby improving the doctor's ability to interpret and analyze the image. Through image segmentation and regional characteristic processing, the method can more accurately identify the region of interest, thereby improving the accuracy of gastric mucosa analysis. Through image enhancement and simulation, the method can provide richer image information, including multi-channel color mapping and simulation data, and can provide rich multi-dimensional data to adapt to the gastric mucosal image data collected under different circumstances.

[0082] Preferably, step S1 specifically includes:

[0083] Step S11: collecting gastric mucosa images through a high-resolution sensor to obtain primary gastric mucosa image data;

[0084] Specifically, a medical device such as a high-resolution endoscope or a gastric mucosal scanner is used to acquire images in the patient's stomach to generate high-quality primary gastric mucosal image data, including multiple viewing angles and resolutions. A high-resolution endoscope device is used to acquire images in the patient's stomach. Ten primary gastric mucosal images are acquired, each image including a different viewing angle and resolution. The image acquisition device has the following parameters: a resolution of 1920x1080 pixels, an exposure time set to 1 / 30 seconds, and a white LED light source.

[0085] Step S12: performing standardization processing on the primary gastric mucosa image data to obtain standard gastric mucosa image data;

[0086] Specifically, white balance: corrects the color temperature of the image to ensure that colors are accurately reflected. Brightness / contrast adjustment: adjusts the brightness and contrast of the image to optimize visibility. Geometric correction: corrects the geometric distortion of the image, such as removing perspective distortion. Noise removal: uses noise reduction algorithms, such as mean filtering or wavelet noise reduction, to reduce noise in the image.

[0087] Step S13: performing metadata annotation on the standard gastric mucosa image data to obtain gastric mucosa image data.

[0088] Specifically, tissue recognition: Use computer vision technology or deep learning models to identify different tissue types in the image, such as mucosa, mucus, blood vessels, etc. Regional distribution description: Describe the distribution of different tissues in the image, including their relative position and density. Image acquisition parameters: Record the parameters during image acquisition, such as resolution, exposure time, light source type, etc., to provide information about image acquisition conditions.

[0089] Specifically, tissue recognition: Use deep learning convolutional neural network (CNN) for tissue classification. Train the model to distinguish gastric mucosal tissue from other structures (such as mucus or gastric bubbles) in the image. Perform tissue recognition on 10 standard images.

[0090] Regional distribution description: Based on the tissue recognition results, calculate the distribution percentage and relative position of different tissue types (mucosa, mucus, blood vessels, etc.) in the image. For example, in the first image, the mucosal tissue accounts for 60% of the total area and is distributed in the upper left corner, and the mucus is distributed in the lower right corner. Image acquisition parameters: Record the parameters during the image acquisition process, such as resolution, exposure time, light source type, etc. For example, the resolution of the first image is 1920x1080 pixels, the exposure time is 1 / 30 second, and a white LED light source is used.

[0091] In the present invention, by collecting images with a high-resolution sensor, detailed and high-quality gastric mucosal image data can be obtained, providing a reliable basis for subsequent analysis. Standardization ensures the consistency of the image, helps to eliminate changes caused by different acquisition equipment or environmental conditions, and makes the data easier to compare and analyze with other data. Metadata annotation enriches the additional information of the image, which helps to better manage, retrieve and identify the image.

[0092] Preferably, step S13 is specifically:

[0093] Step S131: performing tissue recognition on the standard gastric mucosa image data to obtain tissue recognition data;

[0094] Specifically, a deep learning convolutional neural network (CNN) is used for tissue classification. The model is trained to distinguish gastric mucosal tissue from other structures (such as mucus or gas bubbles) in images. The model can be trained based on labeled training data to automatically identify different tissue types.

[0095] Step S132: generating a tissue distribution description for the standard gastric mucosa image data according to the tissue identification data to obtain tissue distribution description data;

[0096] Specifically, based on the tissue recognition results, the distribution percentage or density of different tissue types in the image is calculated, covering the tissue distribution description of the entire image or a specific area.

[0097] Step S133: in the process of gastric mucosa image acquisition, image acquisition parameter data is acquired;

[0098] Specifically, various parameters are obtained from an image acquisition device (such as an endoscope system), including resolution, exposure time, light source type, shooting time, etc.

[0099] Step S134: metadata annotation is performed on the standard gastric mucosa image data according to the image acquisition parameter data and the tissue distribution description data to obtain gastric mucosa image data.

[0100] Specifically, the tissue identification data and tissue distribution description data are combined with the image acquisition parameter data to create metadata containing various tags. Tags can include tissue type, distribution information, image quality indicators (such as contrast, saturation), etc., which are stored in the metadata of the image for subsequent analysis and processing.

[0101] Specifically, a deep learning convolutional neural network (CNN) was used for tissue classification. First, a large amount of labeled training data was collected, which included standard gastric mucosal images and the tissue type labels associated with them, which could be mucosa, mucus, bubbles, etc. Next, a CNN model was trained to automatically identify different tissue types by learning features in the image. The training process covered thousands of images to ensure that the model could accurately classify in various situations. Once the model training was completed, it was applied to standard gastric mucosal image data. The model identified the tissue type in each image as mucosa, mucus, or other. These results constituted tissue recognition data, which described the distribution of different tissue types in the image. Based on the tissue recognition data, the medical image analyst calculated the distribution of each tissue type in the image. For example, the percentage of mucosal tissue in the image, as well as the location and density of mucus and bubbles, could be determined. This information constituted tissue distribution description data. In image acquisition devices (such as endoscope systems), the system records various image acquisition parameters, including resolution, exposure time, light source type, shooting time, etc. The tissue identification data, tissue distribution description data and image acquisition parameter data are combined to create metadata for the image, which adds various tags to each image for subsequent analysis and processing. For example, the metadata includes the following information: Tissue type: labeled as mucosa, mucus, etc. Tissue distribution information: describes the relative position and density of mucosa, mucus, etc. Image quality indicators: including contrast, saturation, etc. Image acquisition parameters: including resolution, exposure time, etc. The present invention enables users to better understand the content and quality of the image, and supports subsequent Helicobacter pylori detection and gastric mucosal analysis.

[0102] The present invention improves the efficiency and accuracy of image analysis through automated tissue recognition, and can more accurately locate and describe tissue structures in images. Generating tissue distribution description data helps users better understand the location and characteristics of different tissues in the image, thereby providing more information. Metadata annotation makes image data more informative, helps medical experts better understand and use images, and improves the efficiency of data management and utilization. Associating image acquisition parameters with tissue distribution descriptions helps users better understand the background information of the image, which is helpful for image analysis.

[0103] Preferably, step S2 specifically includes:

[0104] Step S21: performing multi-scale image segmentation on the gastric mucosa image data to obtain gastric mucosa image segmentation data;

[0105] Specifically, multiscale filtering: Use multiscale filtering methods, such as Gaussian pyramid or wavelet transform, to generate image pyramids of different scales. Then, apply image segmentation algorithms, such as threshold segmentation, region growing, or deep learning-based segmentation, at each scale to obtain multi-scale segmentation results. Semi-supervised learning: Use labeled image data and unlabeled data for semi-supervised learning to improve segmentation model performance. For example, unlabeled image data can be used together with labeled data to train an image segmentation model using a semi-supervised deep learning method. Adaptive scale selection: Select the appropriate scale level for image segmentation based on the local characteristics and structure of the image to avoid over- or under-segmentation, which is achieved through local feature analysis or image texture analysis.

[0106] Specifically, there is a gastric mucosa image that needs to be segmented at multiple scales. The Gaussian pyramid is used as a multiscale filtering method, and then the region growing algorithm is applied for segmentation. The following is a detailed description of the process: Multiscale filtering: First, start with the original gastric mucosa image, which is a high-resolution image that contains different structures and textures of the gastric mucosa. Create a Gaussian pyramid that includes multiple scale levels. Each level is obtained by downsampling the image of the previous level, producing images of different scales. The purpose of the Gaussian pyramid is to provide images of multiple resolutions for segmentation at different scales. Region growing segmentation: Select an appropriate scale level, which is usually selected based on the size and texture of the structure in the image. At this level, the region growing algorithm begins to be applied. A seed point is selected as the starting point, which is usually located in a region of interest in the image and is automatically selected according to predefined criteria. Starting from the seed point, the region growing algorithm merges adjacent pixels into the same region according to some rules, such as based on the gray value similarity, distance, etc. between pixels. By continuously adding pixels to the region until the stop condition is reached, a segmented region is obtained, representing a part of the structure in the image, such as the mucosal region. Multi-scale result merging: Repeat the above steps and select different scale levels to obtain multiple segmentation results, each of which corresponds to image segmentation at a different scale. Finally, merge the segmentation results of multiple scales, considering the weight of each result, to obtain the final multi-scale segmentation result.

[0107] Step S22: performing regional characteristic processing according to the gastric mucosa image segmentation data to obtain regional characteristic data.

[0108] Specifically, color feature extraction: extract color information from each segmented region, such as color histogram or color texture features, which can be used to distinguish the color characteristics of different regions. Texture feature extraction: perform texture feature analysis on each segmented region, such as local texture direction, texture density, etc., which helps to identify texture differences in the image. Shape feature extraction: measure the shape characteristics of each region, such as area, perimeter, compactness, etc. These characteristics can be used to distinguish structures of different shapes. Spatial relationship feature extraction: analyze the spatial relationship between segmented regions, such as relative position, proximity, etc., which helps to capture the topological information between structures.

[0109] Specifically, a gastric mucosa image that has been segmented into multiple regions is used, and different feature data is extracted from each region. Color feature extraction: For each segmented region, a color histogram is calculated, including the color distribution of each region, such as a histogram of the RGB color channel or a hue distribution in the HSV color space. For example, one region has a dominant red hue, while another region is green or blue. Texture feature extraction: Texture feature extraction methods such as Local Binary Pattern (LBP) or Gray Level Co-occurrence Matrix (GLCM) are used to analyze the texture characteristics of each region and capture the texture differences between different regions in the image. For example, some regions have a rough texture, while other regions have a smooth texture. Shape feature extraction: Measuring the shape characteristics of each region, including area, perimeter, compactness, etc., helps to describe the shape of the region. For example, a larger region has a larger area and perimeter, while a small region has a smaller area and perimeter. Spatial relationship feature extraction: Analyzing the spatial relationship between segmented regions, such as their relative position and proximity, helps to capture the topological information between structures. For example, determining whether two regions are adjacent or overlap each other.

[0110] The multi-scale image segmentation in the present invention helps to separate different tissues and structures in the image, providing a clear basis for further analysis. Regional characteristic processing provides users with more detailed information about different tissues or structures by extracting characteristic information of each region, which helps more accurate analysis and diagnosis. The present invention can deeply understand gastric mucosal images, thereby providing users with more information and promoting more accurate image analysis.

[0111] Preferably, step S21 is specifically:

[0112] Step S211: performing multi-scale filtering on the gastric mucosa image data to obtain multi-scale gastric mucosa image data;

[0113] Specifically, Gaussian pyramid filtering: The original gastric mucosal image is decomposed into a Gaussian pyramid to generate images of different scales. A Gaussian filter is applied to each scale image to smooth the image and reduce noise. Wavelet transform: The image is decomposed into multiple scales using wavelet transform to obtain coefficients containing information of different scales. Coefficients of a specific scale can be selected for segmentation.

[0114] Specifically, a high-resolution gastric mucosa image. First, a Gaussian filter is applied to the original image. Specifically, a Gaussian kernel with a small standard deviation is used to blur the image, and then the image is downsampled to generate a low-resolution image of the first layer (the bottom layer). Next, the Gaussian filter is applied to the original image again, but this time a Gaussian kernel with a larger standard deviation is used, and then it is downsampled again to generate a low-resolution image of the second layer. And so on, a series of low-resolution images are generated, each with different degrees of smoothness, and these images together form a Gaussian pyramid, each layer of which can be used for image segmentation and feature extraction at different scales.

[0115] Specifically, a high-resolution gastric mucosa image. First, wavelet transform is applied to the original image to decompose it into wavelet coefficients of multiple scales. The wavelet coefficients of each scale represent the detailed information of the image at that scale. Wavelet coefficients of a specific scale are selected for image segmentation. For example, high-frequency wavelet coefficients can be used to detect edges and texture details in an image, while low-frequency wavelet coefficients can be used to represent the overall structure of the image. The system can capture different features of the image at different scales to help image segmentation.

[0116] Step S212: performing texture feature extraction and gradient feature extraction on the multi-scale gastric mucosa image data to obtain gastric mucosa image texture feature data and gastric mucosa image gradient feature data;

[0117] Specifically, texture features: Calculate texture features for each scale of the image region, such as Local Binary Pattern, Histogram of Directed Gradients (HOG), etc. These features are used to describe the texture information of the image. Gradient features: Calculate the gradient information of the image at each scale, such as Sobel gradient or Scharr gradient. The gradient represents the edge and structure information in the image.

[0118] Step S213: performing scale selection according to the gastric mucosa image texture feature data and the gastric mucosa image gradient feature data to obtain scale selection data;

[0119] Specifically, scale selection based on texture and gradient features: By analyzing the texture and gradient features at different scales, the appropriate scale level is selected to ensure that the segmentation can capture the structure and edges in the image. Thresholding or adaptive selection: According to the statistical information of texture and gradient features or specific criteria, the optimal scale required for segmentation is selected.

[0120] Specifically, texture feature analysis: First, the texture features are calculated for each image region at each scale, such as Local Binary Pattern (LBP) or other texture descriptors, which reflect the texture information in the image, such as details, texture density, etc. Gradient feature analysis: At the same time, the gradient features, such as Sobel gradient or Scharr gradient, are calculated for each image region at each scale to detect the edge and structure information in the image. Scale selection strategy: Based on the analysis of texture and gradient features, the appropriate scale level is selected to perform image segmentation. Scale selection can adopt one of the following strategies: a. Selection based on texture features: The scale is selected by analyzing the changes in texture features. For example, if the texture features at a certain scale are more obvious, then the scale can be selected as the scale for segmentation. b. Selection based on gradient features: The appropriate scale is selected by checking the gradient features at different scales. If the edge information is clearer at a certain scale, then the scale can be selected. c. Threshold selection: The scale is selected based on the statistical information of texture and gradient features or a predefined threshold. For example, a threshold for texture features can be set. If the texture features at a certain scale exceed the threshold, the scale is selected for segmentation. d. Adaptive selection: Use an adaptive algorithm to select the scale and dynamically determine the optimal scale based on the image characteristics at each scale, such as through machine learning methods, information entropy maximization and other technologies.

[0121] Assume that there is a gastric mucosa image. After multi-scale filtering, three image pyramids of different scales are generated: small scale, medium scale and large scale. For each scale, texture features and gradient features are calculated. At the small scale, the texture features of the image are more obvious, but the edge information is not clear enough. At the medium scale, the texture features and edge information of the image are both expressed to a certain extent. At the large scale, the texture features of the image are weaker, but the edge information is clear.

[0122] The system uses an adaptive selection strategy to dynamically adjust the scale according to the weights of texture features and gradient features. If the weight of the texture feature is high, a small scale is selected; if the weight of the gradient feature is high, a large scale is selected. In this way, the system can flexibly select the most suitable scale according to the specific characteristics of the image to perform the best image segmentation. The scale selection strategy will automatically adjust according to different images to adapt to different gastric mucosal images.

[0123] Step S214: performing edge segmentation on the multi-scale gastric mucosa image data according to the scale selection data to obtain gastric mucosa image segmentation data.

[0124] Specifically, Canny edge detection: Apply the Canny edge detection algorithm to segment the image based on the scale selection data. The Canny algorithm can identify strong edges in the image. Threshold segmentation: Use a threshold to separate the edges in the image from the background, and adjust the threshold based on the scale selection data to adapt to features of different scales.

[0125] The present invention introduces scale selection data, and adaptively selects a suitable scale level according to the local texture and gradient features of the image. Traditional methods are difficult to adapt to multiple structures of different scales and complexities at the same time, resulting in under-segmentation or over-segmentation problems. The present invention avoids the scale selection problem in the traditional method and helps to better adapt to diverse structures. By extracting the texture features and gradient features of the image, the present invention can consider the image information more comprehensively, and improve the precision and accuracy of the segmentation. The scale selection process is automated, does not need to rely on manual parameter selection, reduces the interference of subjective factors, and improves the intelligence of the algorithm. By performing multi-scale filtering on the gastric mucosal image data to generate multi-scale gastric mucosal image data, it is helpful to capture the features and structures of different scales in the image, and a variety of information from microscopic to macroscopic can be effectively included in the analysis, which improves the comprehensiveness of image segmentation. Extracting the texture features and gradient features of the gastric mucosal image data makes the segmentation process more informative. The texture features can capture the details and texture information of the image, while the gradient features can help determine the edges and contours. The combination of the two improves the accuracy of segmentation. Using the texture features and gradient features of gastric mucosal images for scale selection makes the segmentation process more adaptive, helps overcome the limitation of relying on fixed scales in traditional methods, and improves the accuracy of segmentation. Edge segmentation of multi-scale gastric mucosal image data based on scale selection data helps to accurately locate the structure and regional boundaries in gastric mucosal images and provide more accurate segmentation results.

[0126] Preferably, step S22 is specifically:

[0127] Performing color feature extraction, texture feature extraction, shape feature extraction, and spatial relationship feature extraction on the gastric mucosa image segmentation data to obtain gastric mucosa image color feature data, gastric mucosa image texture feature data, gastric mucosa image shape feature data, and gastric mucosa image spatial relationship feature data;

[0128] Specifically, color feature extraction includes: Histogram statistics: Perform color histogram statistics on the pixel values ​​in each segmented area to obtain the distribution information of different color channels (such as RGB or HSV). Color space conversion: Convert an image from one color space to another, such as from RGB to Lab color space, to extract different color features.

[0129] Texture feature extraction includes: GLCM feature: Calculate the Gray-Level Co-occurrence Matrix features of each region, including energy, contrast, uniformity, etc., to describe texture information. Gabor filter: Apply Gabor filter to extract the texture characteristics of the region, which is used to capture textures of different directions and scales.

[0130] Shape feature extraction includes: Contour features: Extract contour features of each segmented region, such as perimeter, area, shape factor, etc., to describe the shape of the region. Edge detection: Perform edge detection on each region to obtain edge-related shape features.

[0131] Spatial relationship feature extraction: Relative position feature: Calculate the relative position relationship between different regions, such as distance, relative direction, etc. Connectivity analysis: Analyze the connectivity and topological relationship between regions to identify specific spatial patterns.

[0132] Performing feature selection on the gastric mucosa image color feature data, the gastric mucosa image texture feature data, the gastric mucosa image shape feature data, and the gastric mucosa image space relationship feature data to obtain color feature selection data, texture feature selection data, shape feature selection data, and space relationship feature selection data;

[0133] Specifically, feature selection: Based on information gain: Use information theory methods to select the most informative color, texture, shape and spatial relationship features to reduce redundant information. Principal component analysis (PCA): Convert multiple features into fewer dimensions through PCA dimensionality reduction technology, retaining the main information.

[0134] The color feature selection data, the texture feature selection data, the shape feature selection data and the spatial relationship feature selection data are fused to obtain the regional feature data.

[0135] Specifically, feature fusion: Weight assignment: assign weights to each feature so that they can be combined into the final regional feature data according to their relative importance. Feature combination: combine different feature data, which can be simple splicing or complex mathematical operations, to generate comprehensive regional feature data.

[0136] Specifically, feature selection: Color feature: used to describe the color information of each region, such as color histogram. Texture feature: used to describe the texture information of each region, such as local binary pattern (LBP) feature. Shape feature: used to describe the shape information of each region, such as area, perimeter, etc. Spatial relationship feature: used to describe the relationship between each region and the surrounding areas, such as relative position, proximity, etc. Weight assignment: assign weights to each feature to reflect its relative importance in the final regional feature, which can be determined by the knowledge of domain experts, data analysis or experiments. Assume that the color feature is the most important in image analysis, so a higher weight is assigned to the color feature, such as 0.4. Texture features are also important and are assigned a weight of 0.3. Shape features are relatively minor and are assigned a weight of 0.2. Spatial relationship features have different importance in specific applications and are assigned a weight of 0.1. Feature fusion: In order to generate the final regional feature data, the data of each feature is multiplied by its corresponding weight and combined together. For example, for a specific area, the color feature score is 0.8, the texture feature score is 0.6, the shape feature score is 0.3, and the spatial relationship feature score is 0.5. The following formula can be used for fusion: Final regional feature = (0.4*color feature) + (0.3*texture feature) + (0.2*shape feature) + (0.1*spatial relationship feature).

[0137] In the present invention, by extracting multidimensional characteristics such as color, texture, shape and spatial relationship from the gastric mucosal image segmentation data, various information of the image can be fully captured, including color distribution, texture characteristics, shape contours and spatial relationships between objects, so as to more comprehensively describe the characteristics of each segmented area. After extracting the multidimensional characteristics, screening and combining the data according to the characteristic selection helps to optimize the combination of characteristics, select the most relevant and useful characteristics, reduce redundant information and improve the accuracy of segmentation. By fusing the color, texture, shape and spatial relationship characteristic data after characteristic selection, the information between different characteristics can be correlated to provide more comprehensive regional characteristic data, which helps to more accurately describe each segmented area.

[0138] Preferably, the gastric mucosa image enhancement data includes first gastric mucosa image enhancement data and second gastric mucosa image enhancement data, and step S3 is specifically as follows:

[0139] Perform weighted reset reliability calculation based on regional characteristic data to obtain regional characteristic weighted reset reliability data;

[0140] Specifically, based on the regional characteristic data, weighted confidence data of each region is calculated, such as estimated based on statistical information, a machine learning model, or a rule definition.

[0141] A set of gastric mucosa images are analyzed, each image is segmented into multiple regions, and it is hoped that a weighted confidence is calculated for each region to determine its importance in image enhancement. Regional feature data: First, various feature data such as color features, texture features, shape features, and spatial relationship features have been extracted from each region, which are used to describe the different characteristics of each region. Weighted confidence calculation: It is hoped that the weighted confidence data of each region will be calculated to determine which regions need more attention and enhancement, and which can remain unchanged or reduce processing. For this example, statistical information and machine learning methods will be used to calculate weighted confidence: a. Statistical information method: Statistical information such as mean, variance, median, etc. are calculated for each feature (color, texture, shape, spatial relationship). For each region, these statistics are aggregated to form a feature vector. Then, clustering algorithms (such as K-means clustering) can be used to divide the regions into different groups, each with similar feature vectors. Finally, weighted confidence is assigned to each region according to the group to which it belongs. For example, regions located in a group similar to the gastric mucosa have higher weighted confidence because they are more representative. b. Machine learning method: Use the feature data as input features and the weighted confidence as the target variable to build a machine learning model, such as a regression model or a classification model. Use labeled image data to train the model to learn how to predict the weighted confidence based on the feature data. After training, the model can be applied to new images to predict the weighted confidence for each region.

[0142] When it is determined that the regional characteristic weighted reset confidence data is less than or equal to the preset weighted reset confidence threshold data, local noise reduction strategy data is generated according to the regional characteristic data, and the gastric mucosa image data is subjected to noise reduction processing according to the local noise reduction strategy data to obtain first gastric mucosa image enhancement data;

[0143] Specifically, local noise reduction strategy generation, based on the regional characteristic data, generates a local noise reduction strategy for each region, for example, median filtering, wavelet noise reduction or filtering methods based on neighborhood statistical information can be selected. Noise reduction processing, applying the local noise reduction strategy to the gastric mucosal image data to reduce noise and improve image quality.

[0144] When it is determined that the regional characteristic weighted reset confidence data is greater than the preset weighted reset confidence threshold data, high-pass filtering is performed on the gastric mucosa image data according to the regional characteristic data to obtain gastric mucosa image high-pass filtered data;

[0145] Specifically, high-pass filtering uses a high-pass filter (such as Laplace, Sobel, etc.) to process the gastric mucosal image to enhance the edges and details of the image.

[0146] Performing spatial filtering on the high-pass filtered data of the gastric mucosa image to obtain spatial filtered data of the gastric mucosa image;

[0147] Specifically, spatial filtering applies different spatial domain filters (such as mean, Gaussian filters) to smooth the image to reduce noise and details to obtain a smoother image.

[0148] Dynamic range compression is performed on the spatially filtered data of the gastric mucosa image to obtain second gastric mucosa image enhanced data.

[0149] Specifically, dynamic range compression is performed on the spatially filtered data of the gastric mucosal image to enhance the contrast and details of the image, which is achieved by applying histogram equalization or other contrast enhancement techniques.

[0150] Specifically, an adaptive approach will be used to perform dynamic range compression to ensure that images in different regions are appropriately enhanced, rather than a global approach. Step 1: Local contrast estimation: The spatially filtered data of the gastric mucosal image is segmented to divide the image into multiple small blocks or windows, which can be overlapping or non-overlapping, depending on the requirements. For each window, a local contrast estimate is calculated, which is achieved by calculating the ratio of the standard deviation to the mean of the pixels within the window. The local contrast estimate reflects the contrast level of the image within each window. Step 2: Dynamic range compression: For each window, the degree of dynamic range compression applied is determined based on its local contrast estimate. A higher local contrast estimate requires a stronger dynamic range compression, while a lower local contrast estimate requires a lighter compression. An adaptive dynamic range compression algorithm, such as histogram equalization, is applied to the pixels within each window, but is adjusted based on the contrast estimate of the window to enhance the contrast of the image, but avoid over-enhancement, thereby retaining more details. Processing is performed window by window to ensure that each region obtains appropriate dynamic range compression to optimize the visual quality of the image. Step 3: Recombining the image, after processing all windows, the windows with dynamic range compression are recombined into the final image, which can be overlapped or spliced ​​according to the position information between windows to ensure a smooth transition. The final second gastric mucosa image enhancement data has locally adaptive dynamic range compression to enhance the contrast and details of the image while maintaining the appropriate brightness level of each area.

[0151] In the present invention, when the regional characteristic weighted reset confidence data is less than or equal to the preset weighted reset confidence threshold data, a local noise reduction strategy is used to process the image, which helps to remove noise and unnecessary details and improve the quality and readability of the image. When the regional characteristic weighted reset confidence data is greater than the preset weighted reset confidence threshold data, a high-pass filter is used to process the image, and the high-pass filter helps to enhance the edges and details in the image, making the target structure more clearly visible. Spatial filtering and dynamic range compression are performed on the high-pass filtered image, which helps to further improve the quality and contrast of the image, making the image more visually attractive and highlighting important features. Two different gastric mucosal image enhancement data are generated, the first gastric mucosal image enhancement data is mainly used for local noise reduction, and the second gastric mucosal image enhancement data is mainly used for edge and detail enhancement. The dual enhancement strategy can better meet the needs of different application scenarios.

[0152] Preferably, step S4 is specifically:

[0153] Step S41: acquiring region of interest data, and segmenting the gastric mucosa image enhancement data according to the region of interest data to obtain gastric mucosa region of interest image data;

[0154] Specifically, based on the gastric mucosal image and related information, the location and boundary of the region of interest are determined, for example, a region with specific features (such as specific features determined to be abnormal. For each region, the anomaly detection model usually outputs a confidence score indicating whether the region is abnormal. The confidence score can be a probability, a score, or other quantitative indicators. A higher score indicates that the region is abnormal, and a lower score indicates that it is normal.) Based on the boundary information of the region of interest, the image data of the region of interest is extracted from the gastric mucosal image enhancement data.

[0155] Step S42: performing multimodal color mapping on the gastric mucosa region of interest image data to obtain region of interest image color mapping data;

[0156] Specifically, multimodal color mapping is applied to the image data of the region of interest to map different information (such as structure, blood flow, metabolism) to different color channels. For example, the structural information can be mapped to grayscale, the blood flow information can be mapped to red, and the metabolic information can be mapped to green.

[0157] Step S43: performing contrast enhancement on the image color mapping data of the region of interest to obtain enhanced image data of the region of interest;

[0158] Specifically, based on the characteristics of the image of the region of interest, contrast enhancement techniques are applied to increase the contrast and details of the image to make the abnormal area more prominent. If there are specific structures or features in the image of the region of interest, enhancement can be performed based on these features. For example, if the system determines that the abnormal area usually has a certain texture feature, a filter or enhancement method can be applied to highlight this texture.

[0159] Step S44: sharpening the image enhancement data of the region of interest to obtain sharpened image data of the region of interest;

[0160] Specifically, the image enhancement data of the region of interest is sharpened to enhance the details and clarity of the image, which can be achieved by using a sharpening filter or an enhancement algorithm.

[0161] Step S45: mapping the region of interest image sharpening data to preset pseudo color region image data, thereby obtaining region of interest image pseudo color data;

[0162] Specifically, the region of interest image sharpening data is mapped to a preset pseudo color map to make different information easier to distinguish. For example, the sharpened image is mapped to a color image to more clearly display structure, blood flow and metabolic information.

[0163] Specifically, a region of interest image, this image has been sharpened to enhance structural information, but it is still a grayscale image. Now we want to map the sharpened image to a pseudo-color image to highlight different information. First, we need to define a pseudo-color mapping, which can be a color mapping table that maps different grayscale values ​​to different colors. For example, low grayscale values ​​can be mapped to blue, medium grayscale values ​​to green, and high grayscale values ​​to red. Next, the sharpened region of interest image is paired with the pseudo-color mapping table, which is implemented through a simple lookup table operation, where each grayscale pixel value is replaced with the corresponding pseudo-color value. Finally, a pseudo-color image is obtained, in which different structures and information are visualized in color, making it easier for users to distinguish different regions and features in the image.

[0164] Step S46: Perform three-dimensional model simulation on the pseudo-color data of the region of interest image to obtain simulated data of the region of interest for performing gastric mucosa image analysis.

[0165] Specifically, based on the pseudo-color data of the region of interest image, a three-dimensional model simulation is performed to generate a visual three-dimensional model for further gastric mucosal image analysis and detection of targets such as Helicobacter pylori.

[0166] Specifically, a pseudo-color image that contains structural and feature information of the area of ​​interest, the system now creates a 3D model to analyze this information in more detail. First, the pseudo-color image is converted into 3D point cloud data, which is achieved by extracting depth information from the pseudo-color image and pairing it with color information. 3D coordinates are assigned to each pixel, thus creating a point cloud. Next, the point cloud data is used to build a 3D model, which is achieved by calculating the relationship between the point clouds and the topological information in the point clouds. A variety of 3D modeling techniques, such as structured light scanning or stereo vision, can be used to build the model. Finally, a visual 3D model is obtained that contains the structure and features of the area of ​​interest. This 3D model can be used for more in-depth analysis, such as detecting the location of Helicobacter pylori or measuring the size and shape of the structure.

[0167] The present invention obtains region of interest data, and segments the gastric mucosal image enhancement data based on these data, which can accurately extract specific areas of concern to doctors or analysts, help focus on important areas, save time and improve efficiency. The multimodal color mapping strategy can better highlight the various information in the region of interest, so that different information can be distinguished in visualization, which helps doctors understand the various features in the image more clearly. Contrast enhancement and sharpening of the region of interest image data helps to enhance the details and edges in the image, making the region of interest clearer and more prominent. Mapping the region of interest image sharpening data to pseudo-color region image data can provide more vivid and intuitive visualization results, which helps to more easily identify and analyze different regions. The three-dimensional model simulation performed in the present invention can provide region of interest simulation data with more depth and stereoscopic sense, which helps users to more comprehensively understand and analyze the structure and features in the image.

[0168] Preferably, step S42 is specifically:

[0169] Performing structural recognition on the gastric mucosa region of interest image data to obtain region of interest structure type data, wherein the region of interest structure type data includes region of interest blood vessel data, region of interest structure data, and region of interest metabolic data;

[0170] Specifically, first, the image data of the gastric mucosa region of interest are subjected to structure recognition analysis to determine different structural types, including blood vessels, tissue structures, and metabolic areas, which are achieved through image processing and segmentation techniques to identify and separate different structural components.

[0171] According to the vascular data, structural data and metabolic data of the region of interest, different channels of the gastric mucosa region of interest image data are mapped respectively to generate multi-channel region of interest image color mapping data.

[0172] Perform green channel mapping on the gastric mucosa region of interest image data according to the region of interest blood vessel data to obtain first region of interest image color mapping data;

[0173] Specifically, based on the region of interest blood vessel data, green channel mapping processing is performed to highlight the blood vessel structure, and first region of interest image color mapping data is generated.

[0174] Performing red channel mapping on the gastric mucosa region of interest image data according to the region of interest structure data to obtain second region of interest image color mapping data;

[0175] Specifically, based on the region of interest structure data, red channel mapping processing is performed to highlight the tissue structure, and second region of interest image color mapping data is generated.

[0176] Performing blue channel mapping on the gastric mucosa region of interest image data according to the region of interest metabolism data to obtain third region of interest image color mapping data;

[0177] Specifically, based on the metabolic data of the region of interest, blue channel mapping processing is performed to highlight the metabolic region, and third region of interest image color mapping data is generated.

[0178] The first region of interest image color mapping data, the second region of interest image color mapping data and the third region of interest image color mapping data are synthesized to obtain the region of interest image color mapping data.

[0179] Specifically, the first, second and third region of interest image color mapping data are synthesized to generate multi-channel region of interest image color mapping data, wherein different color channels respectively reflect different structures and features to provide richer information.

[0180] In the present invention, by performing structural recognition on the image data of the region of interest, different types of structures, such as blood vessels, structures and metabolism, can be distinguished. Then, these types of structures are mapped to different color channels, such as green, red and blue channels. Multimodal color mapping helps users to more clearly distinguish and understand different structures and information in the image. By using different channels for mapping, specific types of structural information can be highlighted. For example, mapping blood vessels to the green channel makes them more eye-catching in the image. Mapping structural information to the red channel and metabolic information to the blue channel helps users to better distinguish and analyze the characteristics of different structures. Conventional methods usually use a single image processing technology, such as segmentation or enhancement, which does not make full use of the multi-channel information of the image, resulting in the omission of important structures and features, limiting the user's comprehensive understanding of the image. The present invention uses multi-channel color mapping to provide a variety of visual information to help users more easily identify the features of the region of interest. The data of each channel contains different information, and the comprehensive use can improve the accuracy of data analysis. By synthesizing the mapping data of different channels into an image, a more comprehensive visualization result can be provided, which helps users see the relationship between different structures in one image and more comprehensively understand the features of the region of interest.

[0181] Preferably, the present application also provides a gastric mucosa image analysis system for Helicobacter pylori detection, which is used to execute the gastric mucosa image analysis method for Helicobacter pylori detection as described above, and the gastric mucosa image analysis system for Helicobacter pylori detection comprises:

[0182] A gastric mucosa image data acquisition module, used for acquiring gastric mucosa image data;

[0183] A regional characteristic processing module is used to perform image segmentation on the gastric mucosa image data to obtain gastric mucosa image segmentation data, and perform regional characteristic processing based on the gastric mucosa image segmentation data to obtain regional characteristic data;

[0184] An image enhancement module, used for performing image enhancement on gastric mucosa image data according to regional characteristic data to obtain gastric mucosa image enhancement data;

[0185] The gastric mucosa image analysis module is used to obtain the data of the region of interest, and simulate the gastric mucosa image enhancement data according to the data of the region of interest to obtain the simulated data of the region of interest for performing gastric mucosa image analysis operations.

[0186] The beneficial effects of the present invention are: through image segmentation and regional characteristic processing, the method can more accurately identify the region of interest, thereby improving the accuracy of gastric mucosa analysis. Through image enhancement and simulation, the method can provide richer image information, including multi-channel color mapping and simulation data, so that users can have a more comprehensive understanding of the condition of the gastric mucosa, so as to better make accurate analysis. The image segmentation, regional characteristic processing and image enhancement steps in the method can be partially or completely automated, thereby reducing the workload of users engaged in related work, improving analysis efficiency, and is very useful for clinical applications that process a large number of gastric mucosal images. The simulation of the region of interest data can be customized according to the specific situation of each user, realizing personalized data analysis, thereby overcoming the differences between different users. By providing more accurate and rich image data, the method helps to make accurate data analysis early and avoids the delay caused by complex biochemical reaction processes.

[0187] Therefore, from any point of view, the embodiments should be regarded as illustrative and non-restrictive, and the scope of the present invention is limited by the attached application documents rather than the above description, and it is intended that all changes falling within the meaning and scope of equivalent elements of the application documents are included in the present invention.

[0188] The above description is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but should conform to the widest scope consistent with the principles and novel features invented herein.

Claims

1. A gastric mucosal image analysis method for Helicobacter pylori detection, characterized in that: The following steps are involved: Step S1: Acquire gastric mucosa image data; Step S2: performing image segmentation on the gastric mucosa image data to obtain gastric mucosa image segmentation data, and performing regional characteristic processing on the gastric mucosa image segmentation data to obtain regional characteristic data; Step S3: performing image enhancement on the gastric mucosa image data according to the regional characteristic data to obtain gastric mucosa image enhancement data; Step S4: acquiring the region of interest data, and simulating the gastric mucosa image enhancement data according to the region of interest data to obtain the region of interest simulation data for performing gastric mucosa image analysis operations; The gastric mucosa image enhancement data includes first gastric mucosa image enhancement data and second gastric mucosa image enhancement data, and step S3 is specifically as follows: Perform weighted reset reliability calculation based on regional characteristic data to obtain regional characteristic weighted reset reliability data; When it is determined that the regional characteristic weighted reset confidence data is less than or equal to the preset weighted reset confidence threshold data, local noise reduction strategy data is generated according to the regional characteristic data, and the gastric mucosa image data is subjected to noise reduction processing according to the local noise reduction strategy data to obtain first gastric mucosa image enhancement data; When it is determined that the regional characteristic weighted reset confidence data is greater than the preset weighted reset confidence threshold data, high-pass filtering is performed on the gastric mucosa image data according to the regional characteristic data to obtain gastric mucosa image high-pass filtered data; Performing spatial filtering on the high-pass filtered data of the gastric mucosa image to obtain spatial filtered data of the gastric mucosa image; Performing dynamic range compression on the gastric mucosa image spatial filtering data to obtain second gastric mucosa image enhancement data; The dynamic range compression is as follows: Segmenting the spatial filtering data of the gastric mucosa image and calculating the local contrast estimation value to obtain the local contrast estimation value data; Performing dynamic range compression according to the local contrast estimation value data to obtain dynamic range compression data; Image reorganization is performed according to the dynamic range compression data to obtain second gastric mucosa image enhancement data.

2. The method according to claim 1, characterized in that Step S1 is specifically as follows: Step S11: collecting gastric mucosa images through a high-resolution sensor to obtain primary gastric mucosa image data; Step S12: performing standardization processing on the primary gastric mucosa image data to obtain standard gastric mucosa image data; Step S13: performing metadata annotation on the standard gastric mucosa image data to obtain gastric mucosa image data.

3. The method according to claim 2, characterized in that Step S13 is specifically as follows: Step S131: performing tissue recognition on the standard gastric mucosa image data to obtain tissue recognition data; Step S132: generating a tissue distribution description for the standard gastric mucosa image data according to the tissue identification data to obtain tissue distribution description data; Step S133: in the process of gastric mucosa image acquisition, image acquisition parameter data is acquired; Step S134: metadata annotation is performed on the standard gastric mucosa image data according to the image acquisition parameter data and the tissue distribution description data to obtain gastric mucosa image data.

4. The method according to claim 1, characterized in that: Step S2 is specifically as follows: Step S21: performing multi-scale image segmentation on the gastric mucosa image data to obtain gastric mucosa image segmentation data; Step S22: performing regional characteristic processing according to the gastric mucosa image segmentation data to obtain regional characteristic data.

5. The method according to claim 4, characterized in that Step S21 is specifically as follows: Step S211: performing multi-scale filtering on the gastric mucosa image data to obtain multi-scale gastric mucosa image data; Step S212: performing texture feature extraction and gradient feature extraction on the multi-scale gastric mucosa image data to obtain gastric mucosa image texture feature data and gastric mucosa image gradient feature data; Step S213: performing scale selection according to the gastric mucosa image texture feature data and the gastric mucosa image gradient feature data to obtain scale selection data; Step S214: performing edge segmentation on the multi-scale gastric mucosa image data according to the scale selection data to obtain gastric mucosa image segmentation data.

6. The method according to claim 4, characterized in that Step S22 is specifically as follows: Performing color feature extraction, texture feature extraction, shape feature extraction, and spatial relationship feature extraction on the gastric mucosa image segmentation data to obtain gastric mucosa image color feature data, gastric mucosa image texture feature data, gastric mucosa image shape feature data, and gastric mucosa image spatial relationship feature data; Performing feature selection on the gastric mucosa image color feature data, the gastric mucosa image texture feature data, the gastric mucosa image shape feature data, and the gastric mucosa image space relationship feature data to obtain color feature selection data, texture feature selection data, shape feature selection data, and space relationship feature selection data; The color feature selection data, the texture feature selection data, the shape feature selection data and the spatial relationship feature selection data are fused to obtain the regional feature data.

7. The method according to claim 1, characterized in that Step S4 is specifically as follows: Step S41: acquiring region of interest data, and segmenting the gastric mucosa image enhancement data according to the region of interest data to obtain gastric mucosa region of interest image data; Step S42: performing multimodal color mapping on the gastric mucosa region of interest image data to obtain region of interest image color mapping data; Step S43: performing contrast enhancement on the image color mapping data of the region of interest to obtain image enhancement data of the region of interest; Step S44: sharpening the image enhancement data of the region of interest to obtain sharpened image data of the region of interest; Step S45: mapping the region of interest image sharpening data to preset pseudo color region image data, thereby obtaining region of interest image pseudo color data; Step S46: Perform three-dimensional model simulation on the pseudo-color data of the region of interest image to obtain simulated data of the region of interest for performing gastric mucosa image analysis.

8. The method according to claim 7, characterized in that Step S42 is specifically as follows: Performing structural recognition on the gastric mucosa region of interest image data to obtain region of interest structure type data, wherein the region of interest structure type data includes region of interest blood vessel data, region of interest structure data, and region of interest metabolic data; Perform green channel mapping on the gastric mucosa region of interest image data according to the region of interest blood vessel data to obtain first region of interest image color mapping data; Performing red channel mapping on the gastric mucosa region of interest image data according to the region of interest structure data to obtain second region of interest image color mapping data; Performing blue channel mapping on the gastric mucosa region of interest image data according to the region of interest metabolism data to obtain third region of interest image color mapping data; The first region of interest image color mapping data, the second region of interest image color mapping data and the third region of interest image color mapping data are synthesized to obtain the region of interest image color mapping data.

9. A gastric mucosal image analysis system for Helicobacter pylori detection, characterized in that: The gastric mucosa image analysis method for Helicobacter pylori detection as claimed in claim 1, wherein the gastric mucosa image analysis system for Helicobacter pylori detection comprises: A gastric mucosa image data acquisition module, used for acquiring gastric mucosa image data; A regional characteristic processing module is used to perform image segmentation on the gastric mucosa image data to obtain gastric mucosa image segmentation data, and perform regional characteristic processing based on the gastric mucosa image segmentation data to obtain regional characteristic data; An image enhancement module, used for performing image enhancement on gastric mucosa image data according to regional characteristic data to obtain gastric mucosa image enhancement data; The gastric mucosa image analysis module is used to obtain the data of the region of interest, and simulate the gastric mucosa image enhancement data according to the data of the region of interest to obtain the simulated data of the region of interest for performing gastric mucosa image analysis operations.

Images