Method and device for identifying helicobacter pylori based on artificial intelligence fluorescence imaging
By using artificial intelligence-based fluorescence imaging technology to dynamically segment and label Helicobacter pylori, the problems of low efficiency and poor consistency in traditional diagnostic methods are solved, achieving efficient and accurate identification and quantitative analysis of Helicobacter pylori.
Patent Information
- Application Number
- CN202511164504.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-08-20
- Publication Date
- 2025-12-02
AI Technical Summary
Traditional methods for diagnosing Helicobacter pylori are cumbersome, time-consuming, slow, and have poor consistency in human interpretation, making efficient identification and quantitative analysis difficult.
An artificial intelligence-based fluorescence imaging method was adopted. Samples were prepared and immunofluorescence staining was performed. Image processing and recognition were carried out using a deep learning model. Combined with U-Net semantic segmentation and GIoU Loss function, Helicobacter pylori was dynamically segmented and labeled, and rectangular box labels and positive probability data were output.
It enables automatic labeling of Helicobacter pylori, improving detection efficiency and identification accuracy, reducing false positives, and ensuring the accuracy and speed of identification of local high-density bacterial communities.
Smart Images

Figure CN121053083A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of artificial intelligence medical detection technology, specifically to a method and apparatus for identifying Helicobacter pylori based on artificial intelligence fluorescence imaging. Background Technology
[0002] Helicobacter pylori infection is caused by a spiral-shaped, microaerophilic bacterium with very demanding growth requirements. It was first successfully isolated from gastric mucosal biopsy tissue of a patient with chronic active gastritis in 1983. Studies indicate that Helicobacter pylori lives in the pylorus of the human stomach and is one of the most common bacterial pathogens; more than half the world's population has been infected with Helicobacter pylori. People are usually infected in childhood, with up to 50% of cases occurring in children under 5 years old. This bacterial infection initially causes chronic gastritis, leading to gastric ulcers and gastric atrophy, and in severe cases, gastric cancer. Early detection of Helicobacter pylori infection and timely and effective eradication of the bacterium with antibiotics are of great significance for the prevention and control of gastric cancer.
[0003] Depending on whether gastroscopy is required, diagnostic testing for Helicobacter pylori can be divided into invasive and non-invasive methods. Non-invasive tests include urea breath testing (UBT), stool antigen testing, and serological testing. Invasive tests include biopsy urease testing, histological examination, and bacterial culture. Among these methods, pathological histological examination is considered the gold standard. Histological examination typically involves gastric mucosal biopsy tissue embedded in paraffin, sectioned, and stained for observation. Staining methods include conventional HE staining and special staining methods such as Giemsa staining and silver staining. However, these staining methods have significant drawbacks, primarily being cumbersome, involving multiple steps, and time-consuming; they are also non-specific staining methods with poor sensitivity for Helicobacter pylori, resulting in low contrast after background staining, making it difficult to detect bacteria, and easily leading to missed diagnoses when the bacterial count is low.
[0004] Diagnostic methods for Helicobacter pylori typically involve qualitative or quantitative analysis of morphological characteristics, combined with staining techniques, to identify the bacteria. Currently, clinical diagnosis mainly relies on traditional staining microscopy and bacterial culture techniques. Traditional staining methods observe the morphological characteristics of Helicobacter pylori in samples under a microscope (such as multiple flagella, blunt ends, and spiral curves) to assist doctors in diagnosing and differentiating Helicobacter pylori. However, these typical diagnostic methods suffer from low detection efficiency, slow diagnostic speed, and poor consistency of human interpretation results. Summary of the Invention
[0005] The purpose of this invention is to provide a method and apparatus for identifying Helicobacter pylori based on artificial intelligence fluorescence imaging, so as to solve the problems mentioned in the background art.
[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for identifying Helicobacter pylori based on artificial intelligence fluorescence imaging, comprising:
[0007] S1: Sample preparation: Prepare cytological samples or tissue sections, and then process them with immunofluorescence staining reagents to obtain fluorescent staining images of Helicobacter pylori.
[0008] S2: Image processing: The fluorescently stained image of Helicobacter pylori is used as the input of the Helicobacter pylori identification neural network model, and the output is to obtain the image of Helicobacter pylori with rectangular box markings;
[0009] S3: Image stitching: Based on the coordinates of the rectangular frame of the Helicobacter pylori image with rectangular frame markings, obtain the coordinate information of Helicobacter pylori in the fluorescently stained image, and stitch the Helicobacter pylori image with rectangular frame markings according to the coordinate information to obtain the stitched image.
[0010] Furthermore, fluorescent staining images of Helicobacter pylori were obtained, including:
[0011] S1.1: Sample processing: Add immunofluorescence staining reagent to the cytology sample or tissue section sample and perform the staining reaction. At the same time, rinse the sample after the reaction and prepare a slide sample by using a coverslip and an anti-fluorescence quenching mounting medium.
[0012] S1.2: Image acquisition: The slide sample is placed on the microscope stage, and fluorescent staining images of Helicobacter pylori are acquired through the microscope.
[0013] Furthermore, the output retrieves images of Helicobacter pylori bacteria marked with rectangular boxes, including:
[0014] S2.1: Segmentation: Based on the distribution characteristics of the Helicobacter pylori fluorescent staining image, the Helicobacter pylori fluorescent staining image is segmented to obtain multiple sub-images. At the same time, the sub-images are numbered and the coordinate information of the sub-images is obtained.
[0015] S2.2: Intelligent recognition: Each sub-image is labeled using a deep learning model, and the labeled sub-images are used as input to the Helicobacter pylori identification neural network model to output sub-images with rectangular box labels.
[0016] Furthermore, obtain the coordinate information of the sub-images, including:
[0017] S2.1.1: Image Analysis: After preprocessing the fluorescently stained image of Helicobacter pylori, it is converted into a grayscale image, and the percentage of positive pixels and grayscale variance are obtained. The high-density areas and hot spots in the grayscale image are determined. At the same time, the standard deviation and average value of the bacterial cell size in the grayscale image are obtained through the semantic segmentation model of U-Net.
[0018] S2.1.2: Image segmentation: Based on the high-density region, hot spot region, standard deviation and average value of bacterial cell size, determine the segmentation method and the final segmentation quantity. At the same time, based on the segmentation method and the final segmentation quantity, segment the preprocessed Helicobacter pylori fluorescent staining image to obtain the segmented sub-images.
[0019] S2.1.3: Sub-image labeling: The sub-images are numbered sequentially, and the coordinate information of the predetermined position of the sub-images is obtained.
[0020] Furthermore, based on the area of the high-density region and the total area, the proportion of the height region is determined; based on the standard deviation and average value of the cell size, the coefficient of variation is determined; and based on the comparison between the proportion of the height region and a preset proportion threshold, and the comparison between the coefficient of variation and a preset variation threshold, the segmentation method is determined, specifically:
[0021] When the proportion of the height region is greater than a preset proportion threshold, the segmentation method is adaptive segmentation; when the proportion of the height region is not greater than the preset proportion threshold and the coefficient of variation is greater than the preset variation threshold, the segmentation method is adaptive segmentation; when the proportion of the height region is not greater than the preset proportion threshold and the coefficient of variation is not greater than the preset variation threshold, the segmentation method is equal segmentation.
[0022] Furthermore, based on the total width and height of the preprocessed Helicobacter pylori fluorescent staining image, and the minimum width and height of the target, the final number of segments is determined, specifically:
[0023]
[0024] Where: N0 is the basic segmentation number, W ove W is the total width of the image. tar H is the target minimum width. ove H represents the total height of the image. tar Where N is the target minimum height, k is the final number of segments, and k is the number of units in the side length of the final segmented mesh. It is a set of positive integers. To round up to the nearest integer.
[0025] Furthermore, the output retrieves the sub-images marked with rectangular boxes, including:
[0026] S2.2.1: Image annotation: Each of the sub-images is preprocessed and data augmented to obtain a processed sub-image. The processed sub-image is used as the input of a deep learning model to output an annotated sub-image.
[0027] S2.2.2: Target detection: The labeled sub-image is used as input to the Helicobacter pylori identification neural network model, and the output is to obtain Helicobacter pylori images with rectangular box labels, wherein the rectangular box labels include the rectangular box coordinates and the positive probability data of Helicobacter pylori.
[0028] Furthermore, the formulas for obtaining the coordinate loss function and classification loss function in the Helicobacter pylori identification neural network model are as follows:
[0029]
[0030] Where: L loc Let L be the coordinate loss function. s.c. For the classification loss function, N loc This represents the number of elements in the labeled sub-image. The rectangular box marks the location of Helicobacter pylori. The coordinate information of the i-th rectangle output by the neural network model for identifying Helicobacter pylori is provided. For manually marked rectangular boxes, identify the coordinate information, F loc (·) represents the GIoU loss function, N a N represents the width of the labeled sub-image. o ρ represents the height of the labeled sub-image. mn The value represents the probability that the Helicobacter pylori marker indicated by the rectangle is positive, where i is the coordinate index of the rectangle marker, m is the width index of the labeled sub-image, and n is the height index of the labeled sub-image.
[0031] Furthermore, based on the coordinates of the rectangular frame and the coordinates of the predetermined orientation of the sub-image, the coordinates of the rectangular frame are converted into coordinate information in the Helicobacter pylori fluorescent staining image, specifically as follows:
[0032]
[0033] Among them: (G) X G Y (S) represents the global coordinates of the target Helicobacter pylori in the original image. X ,S Y (L) represents the coordinates of the origin of the sub-image in the original image. X ,L Y W represents the local coordinates of the target Helicobacter pylori in the subgraph. sH is the pixel width of the sub-image. s This represents the pixel height of the sub-image.
[0034] An apparatus for identifying Helicobacter pylori based on artificial intelligence fluorescence imaging uses any one of the above-described methods for identifying Helicobacter pylori based on artificial intelligence fluorescence imaging.
[0035] Compared with the prior art, the beneficial effects of the present invention are:
[0036] Firstly, this invention uses a neural network model to automatically label Helicobacter pylori, thereby avoiding the subjectivity and time-consuming problems of traditional manual microscopic examination. It can also accurately label the location of bacteria and output positive probability data, thus improving detection efficiency.
[0037] Secondly, this invention dynamically selects the segmentation method based on the proportion of high-density areas in the image and the coefficient of variation of bacterial cell size, thereby ensuring the recognition accuracy of local high-density bacterial colony areas. At the same time, by combining grayscale image analysis and semantic segmentation results, the number of segments is dynamically determined, thereby ensuring that the size of the sub-images is adapted to the bacterial cell distribution characteristics.
[0038] Thirdly, this invention uses the GIoU Loss function as the coordinate loss function and combines it with the classification loss function to suppress false positives. This allows the Helicobacter pylori identification neural network model to adaptively adjust the false positive rate through negative samples during training, thereby improving specificity. Attached Figure Description
[0039] Figure 1 This is a schematic diagram of the method for identifying Helicobacter pylori in this invention;
[0040] Figure 2 This is a fluorescent staining image of Helicobacter pylori in this invention;
[0041] Figure 3 The fragments of Helicobacter pylori in the fluorescent staining image of this invention are... Figure 1 ;
[0042] Figure 4 The fragments of Helicobacter pylori in the fluorescent staining image of this invention are... Figure 2 ;
[0043] Figure 5 The fragments of Helicobacter pylori in the fluorescent staining image of this invention are... Figure 3 ;
[0044] Figure 6 The fragments of Helicobacter pylori in the fluorescent staining image of this invention are... Figure 4 ;
[0045] Figure 7 This is a fluorescent staining image of Helicobacter pylori labeled in this invention;
[0046] Figure 8 This is an image of Helicobacter pylori with a rectangular frame in this invention. Detailed Implementation
[0047] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0048] In the description of this invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.
[0049] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can also refer to the internal connection of two components; and they can refer to a wireless connection or a wired connection. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.
[0050] Furthermore, the technical features involved in the different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.
[0051] refer to Figures 1-8 This embodiment provides a method for identifying Helicobacter pylori based on artificial intelligence fluorescence imaging, which includes the following steps:
[0052] Step S1: Sample Preparation. This involves preparing cytological samples or tissue sections obtained from cell brushes, followed by immunofluorescence staining to obtain fluorescently stained images of Helicobacter pylori. Details are as follows:
[0053] Step S1.1: Sample Processing. This involves gently scraping cytological samples from the gastric mucosa surface using a sterile cell brush, and then evenly spreading the collected cytological samples onto a clean glass slide to obtain corresponding cell brush samples. In this embodiment, gastric mucosal biopsy tissue can also be fixed with 10% neutral formalin and then embedded in paraffin using conventional methods to obtain tissue sections of 3-5 μm. Further, the obtained tissue sections are attached to a glass slide treated with poly-L-lysine and baked at 60°C for 1 hour to obtain corresponding tissue section samples.
[0054] Furthermore, the obtained tissue section samples were sequentially immersed in xylene I, xylene II, anhydrous ethanol, and 95% ethanol, with each immersion in xylene I and xylene II for 10 minutes and in anhydrous ethanol and 95% ethanol for 5 minutes. The immersed tissue section samples were then rinsed under running water for 5 minutes to remove paraffin components.
[0055] In this embodiment, a specific anti-Helicobacter pylori fluorescently labeled antibody reagent containing a fluorescein (e.g., FITC) labeled antibody component is used as the staining reagent. Using a micropipette, 25 μL (approximately 1 drop) of the staining reagent is aspirated and placed in the center of the cell brush / tissue section sample. Simultaneously, a coverslip or spreader is used to evenly distribute the staining reagent throughout the entire sample area, ensuring complete coverage of the specimen surface without air bubbles. The stained sample is then placed in a humidified chamber and incubated at room temperature in the dark for 1-2 minutes.
[0056] Furthermore, after incubation, the sample is gently rinsed with pre-cooled PBS buffer at a 30-45 degree angle for 30 seconds to ensure complete removal of unbound staining reagent. After rinsing, excess liquid is blotted from the edges of the slide with absorbent paper, and a drop of anti-fluorescence quenching mounting medium is added. The coverslip is then slowly placed on the slide at a 45° angle to avoid air bubbles, and any excess mounting medium overflowing from the edges is blotted with absorbent paper.
[0057] Step S1.2: Image Acquisition. The prepared slide sample from step S1.1 is securely placed on the stage of the fully automated fluorescence biological microscope and fixed with slide clips. After locating the sample area in fluorescence mode using the 20× objective lens, the fluorescence signal is searched at low light intensity, and a clear focal plane is obtained by using the coarse / fine adjustment knob.
[0058] In this embodiment, the entire sample area is scanned, and images are simultaneously captured using a research-grade CCD or sCMOS camera with a resolution of at least 1920×1440 pixels, with images taken sequentially at low and high magnification. Specifically, suspicious areas are photographed at low magnification (20×), and typical bacterial morphology is confirmed at high magnification (40×). This yields fluorescently stained images of Helicobacter pylori, such as... Figure 2 As shown.
[0059] Step S2: Image Processing. The fluorescently stained image of Helicobacter pylori obtained in step S1.2 is used as input to the Helicobacter pylori labeling neural network model, and the output is the corresponding image of Helicobacter pylori with a rectangular bounding box. Details are as follows:
[0060] Step S2.1: Segmentation. Based on the distribution characteristics of the Helicobacter pylori fluorescent staining images, the images are segmented to obtain multiple sub-images. These sub-images are then numbered, and their corresponding coordinate information is obtained. Details are as follows:
[0061] Step S2.1.1: Image Analysis. This involves preprocessing the Helicobacter pylori fluorescent staining image obtained in Step S1.2, including image enhancement, size normalization, noise reduction, and color correction, to ensure that the minimum pixel standard of the Helicobacter pylori fluorescent staining image is not lower than 2736×1824 pixels. It is worth noting that the image enhancement, size normalization, noise reduction, and color correction in this embodiment are all conventional image processing methods, and therefore will not be specifically described in this embodiment.
[0062] Furthermore, the preprocessed Helicobacter pylori fluorescent staining image is converted into an 8-bit grayscale image, retaining intensity values from 0 to 255. Simultaneously, the grayscale image is divided into a 10*10 uniform grid, and the percentage of positive pixels within each grid cell is obtained. Based on a comparison between the percentage of positive pixels and a preset threshold range (20%-60%), the image region is divided, specifically as follows:
[0063] When the proportion of positive pixels is less than the lower limit of the preset proportion threshold range (i.e., 20%), the corresponding area is a sparse area. When the proportion of positive pixels is within the preset proportion threshold range (i.e., 20%-60%), the corresponding area is a medium-density area. When the proportion of positive pixels is greater than the upper limit of the preset proportion threshold range (i.e., 60%), the corresponding area is a high-density area.
[0064] In this embodiment, a sliding window is set according to a window size of 100*100 pixels and a step size of 50 pixels. Simultaneously, the grayscale variance within each window is obtained based on the size of the sliding window. Further, based on the grayscale variance within each window, the corresponding mean variance and standard deviation are obtained. A preset difference threshold is then set based on the mean variance and standard deviation, which is the sum of the mean variance and twice the standard deviation.
[0065] Furthermore, the grayscale variance within each window is compared with a preset difference threshold, and hotspot areas are determined based on the comparison results. Specifically:
[0066] When the grayscale variance is greater than a preset difference threshold, the corresponding window area is considered a hotspot area. Conversely, when the grayscale variance is not greater than the preset difference threshold, the corresponding window area is not considered a hotspot area.
[0067] In this embodiment, the converted grayscale image is used as input to the U-Net semantic segmentation model, and the output obtains the size of each bacterial cell. Simultaneously, based on the size of each bacterial cell, the corresponding standard deviation and average size of the bacterial cell are determined.
[0068] Step S2.1.2: Image Segmentation. Based on the high-density area and total area obtained in Step S2.1.1, the corresponding height region proportion (i.e., the ratio between the high-density area and the total area) is determined. Simultaneously, based on the standard deviation and mean of cell size obtained in Step S2.1.1, the corresponding coefficient of variation (i.e., the ratio between the standard deviation and the mean of cell size) is determined. Further, based on the comparison between the height region proportion and a preset proportion threshold (i.e., 30%), and the coefficient of variation and a preset variation threshold (i.e., 25%), the image segmentation method is determined, specifically as follows:
[0069] When the height region's proportion exceeds a preset threshold (30%), the image is segmented using adaptive segmentation. Conversely, when the height region's proportion is not greater than the preset threshold (30%), and the coefficient of variation is greater than the preset variation threshold (25%), the image is segmented using adaptive segmentation. Conversely, when the height region's proportion is not greater than the preset threshold (30%), and the coefficient of variation is not greater than the preset variation threshold (25%), the image is segmented using equal segmentation.
[0070] Furthermore, in this embodiment, multiple Helicobacter pylori bacteria are randomly selected (the specific data can be set according to actual needs; in this embodiment, it is set to 20). At the same time, the width and minor axis diameter of these 20 Helicobacter pylori bacteria are obtained. The width of the bacteria located at the 95th percentile is set as the target minimum width, and the maximum minor axis diameter is set as the target minimum height.
[0071] Furthermore, based on the total width and height of the image, and the minimum width and height of the target, the corresponding number of segments is determined, specifically as follows:
[0072]
[0073] Where: N0 is the basic segmentation number, W ove W is the total width of the image. tar H is the target minimum width. ove H represents the total height of the image. tar Where N is the target minimum height, k is the final number of segments, and k is the number of units in the side length of the final segmented mesh. It is a set of positive integers. To round up to the nearest integer.
[0074] In other words, based on the determined segmentation method and the final number of segments, the preprocessed Helicobacter pylori fluorescent staining image is segmented to obtain multiple segmented sub-images.
[0075] Step S2.1.3: Sub-image labeling. This involves numbering each sub-image sequentially and recording the coordinates of each sub-image's predetermined location (the specific location is not limited in the actual implementation; in this embodiment, it is the upper left corner).
[0076] Specifically, a rectangular coordinate system is established using the top left corner of the preprocessed Helicobacter pylori fluorescent staining image as the origin, with the X-axis pointing to the right and the Y-axis pointing downwards, and the coordinate unit being pixels. For example... Figures 3-6 As shown, the coordinates of sub-image 1 are (0,0), the coordinates of sub-image 2 are (1368,0), the coordinates of sub-image 3 are (0,912), and the coordinates of sub-image 4 are (1368,912).
[0077] Step S2.2: Intelligent Recognition. The sub-image labeled in step S2.1.3 is used as input to the Helicobacter pylori identification neural network model. This model identifies the morphological features of Helicobacter pylori and outputs sub-images marked with rectangular boxes. Specifically:
[0078] Step S2.2.1: Image Annotation. Each sub-image labeled in step S2.1.3 undergoes image preprocessing, including image size normalization, pixel value normalization, and channel normalization. Simultaneously, the preprocessed sub-images are subjected to data augmentation processing, including geometric transformation, photometric adjustment, and elastic deformation, thereby obtaining the processed sub-image. It is worth noting that the image size normalization, pixel value normalization, channel normalization, geometric transformation, photometric adjustment, and elastic deformation in this embodiment are all image preprocessing and data augmentation methods, and therefore will not be specifically described in this embodiment.
[0079] Furthermore, the processed sub-images are used as input to a deep learning model (such as a convolutional neural network model). Through multiple convolutional layers, pooling layers, and fully connected layers within the deep learning model, features in each processed sub-image are identified and classified to obtain labeled sub-images, such as... Figure 7 As shown.
[0080] Step S2.2.2: Object Detection. The labeled sub-images obtained in step S2.2.1 are used as input to the Helicobacter pylori labeling neural network model. The output is an image of Helicobacter pylori with rectangular bounding boxes, such as... Figure 8 As shown. Specifically, the rectangular frame identifier in this embodiment includes the rectangular frame coordinate information and the positive probability data of Helicobacter pylori.
[0081] In this embodiment, the network loss function of the Helicobacter pylori labeling neural network model takes into account the inhibitory effect of negative samples on the false positive rate, supports the addition of negative samples in training, and adjusts the magnitude of the inhibitory effect of negative samples on the false positive rate in model training through hyperparameter control. Furthermore, the inhibitory effect on the false positive rate can be adaptively adjusted according to the negative sample images.
[0082] Specifically, in this embodiment, the network loss function is the sum of the coordinate loss function and the classification loss function. Further, the formulas for obtaining the coordinate loss function and the classification loss function are as follows:
[0083]
[0084] Where: L loc Let L be the coordinate loss function. s.c. For the classification loss function, N loc This represents the number of elements in the labeled sub-image. The rectangular box marks the location of Helicobacter pylori. The coordinate information of the i-th rectangle output by the neural network model for identifying Helicobacter pylori is provided. For manually marked rectangular boxes, identify the coordinate information, F loc(·) represents the GIoU loss function, N a N represents the width of the labeled sub-image. o ρ represents the height of the labeled sub-image. mn The value represents the probability that the Helicobacter pylori marker indicated by the rectangle is positive, where i is the coordinate index of the rectangle marker, m is the width index of the labeled sub-image, and n is the height index of the labeled sub-image.
[0085] Step S3: Image stitching. Based on the rectangular frame coordinate information from step S2.2.2, the coordinate information of Helicobacter pylori in the fluorescently stained image is obtained, and the sub-images are stitched together according to this coordinate information to obtain the stitched image.
[0086] In this embodiment, the rectangular frame coordinate information from step S2.2.2 is converted into coordinate information in the original Helicobacter pylori fluorescent staining image, specifically as follows:
[0087]
[0088] Among them: (G) X G Y (S) represents the global coordinates of the target Helicobacter pylori in the original image. X ,S Y (L) represents the coordinates of the origin of the sub-image in the original image. X ,L Y W represents the local coordinates of the target Helicobacter pylori in the subgraph. s H is the pixel width of the sub-image. s This represents the pixel height of the sub-image.
[0089] Furthermore, based on the converted coordinate information, all sub-images are stitched together to obtain an image of Helicobacter pylori with the Helicobacter pylori identifier and positive probability.
[0090] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended embodiments and their equivalents.
Claims
1. A method for identifying Helicobacter pylori based on artificial intelligence fluorescence imaging, characterized in that, Including: S1: Sample preparation: Prepare cytological samples or tissue sections, and then process them with immunofluorescence staining reagents to obtain fluorescent staining images of Helicobacter pylori. S2: Image processing: The fluorescently stained image of Helicobacter pylori is used as the input of the Helicobacter pylori identification neural network model, and the output is to obtain the image of Helicobacter pylori with rectangular box markings; S3: Image stitching: Based on the coordinates of the rectangular frame of the Helicobacter pylori image with rectangular frame markings, obtain the coordinate information of Helicobacter pylori in the fluorescently stained image, and stitch the Helicobacter pylori image with rectangular frame markings according to the coordinate information to obtain the stitched image.
2. The method for identifying Helicobacter pylori based on artificial intelligence fluorescence imaging according to claim 1, characterized in that, The obtained fluorescent staining images of Helicobacter pylori include: S1.1: Sample processing: Add immunofluorescence staining reagent to the cytology sample or tissue section sample and perform the staining reaction. At the same time, rinse the sample after the reaction and prepare a slide sample by using a coverslip and an anti-fluorescence quenching mounting medium. S1.2: Image acquisition: The slide sample is placed on the microscope stage, and fluorescent staining images of Helicobacter pylori are acquired through the microscope.
3. The method for identifying Helicobacter pylori based on artificial intelligence fluorescence imaging according to claim 1, characterized in that, Output images of Helicobacter pylori bacteria marked with rectangular boxes, including: S2.1: Segmentation: Based on the distribution characteristics of the Helicobacter pylori fluorescent staining image, the Helicobacter pylori fluorescent staining image is segmented to obtain multiple sub-images. At the same time, the sub-images are numbered and the coordinate information of the sub-images is obtained. S2.2: Intelligent recognition: Each sub-image is labeled using a deep learning model, and the labeled sub-images are used as input to the Helicobacter pylori identification neural network model to output sub-images with rectangular box labels.
4. The method for identifying Helicobacter pylori based on artificial intelligence fluorescence imaging according to claim 3, characterized in that, Obtain the coordinate information of the sub-image, including: S2.1.1: Image Analysis: After preprocessing the fluorescently stained image of Helicobacter pylori, it is converted into a grayscale image, and the percentage of positive pixels and grayscale variance are obtained. The high-density areas and hot spots in the grayscale image are determined. At the same time, the standard deviation and average value of the bacterial cell size in the grayscale image are obtained through the semantic segmentation model of U-Net. S2.1.2: Image segmentation: Based on the high-density region, hot spot region, standard deviation and average value of bacterial cell size, determine the segmentation method and the final segmentation quantity. At the same time, based on the segmentation method and the final segmentation quantity, segment the preprocessed Helicobacter pylori fluorescent staining image to obtain the segmented sub-images. S2.1.3: Sub-image labeling: The sub-images are numbered sequentially, and the coordinate information of the predetermined position of the sub-images is obtained.
5. The method for identifying Helicobacter pylori based on artificial intelligence fluorescence imaging according to claim 4, characterized in that, Based on the area of the high-density region and the total area, the proportion of the height region is determined. Based on the standard deviation and average value of the cell size, the coefficient of variation is determined. Simultaneously, based on the comparison between the proportion of the height region and a preset proportion threshold, and the comparison between the coefficient of variation and a preset variation threshold, the segmentation method is determined, specifically: When the proportion of the height region is greater than a preset proportion threshold, the segmentation method is adaptive segmentation; when the proportion of the height region is not greater than the preset proportion threshold and the coefficient of variation is greater than the preset variation threshold, the segmentation method is adaptive segmentation; when the proportion of the height region is not greater than the preset proportion threshold and the coefficient of variation is not greater than the preset variation threshold, the segmentation method is equal segmentation.
6. The method for identifying Helicobacter pylori based on artificial intelligence fluorescence imaging according to claim 4, characterized in that, Based on the total width and height of the preprocessed Helicobacter pylori fluorescent staining image, and the minimum width and height of the target, the final number of segments is determined, specifically: Where: N0 is the basic segmentation number, W ove W is the total width of the image. tar H is the target minimum width. ove H represents the total height of the image. tar The target minimum height, N is the final number of cuts, k is the number of units of the side length of the final cut mesh, and ¢ + It is a set of positive integers. To round up to the nearest integer.
7. The method for identifying Helicobacter pylori based on artificial intelligence fluorescence imaging according to claim 3, characterized in that, The output retrieves the sub-images marked with rectangular boxes, including: S2.2.1: Image annotation: Each of the sub-images is preprocessed and data augmented to obtain a processed sub-image. The processed sub-image is used as the input of a deep learning model to output an annotated sub-image. S2.2.2: Target detection: The labeled sub-image is used as input to the Helicobacter pylori identification neural network model, and the output is to obtain Helicobacter pylori images with rectangular box labels, wherein the rectangular box labels include the rectangular box coordinates and the positive probability data of Helicobacter pylori.
8. The method for identifying Helicobacter pylori based on artificial intelligence fluorescence imaging according to claim 7, characterized in that, The formulas for obtaining the coordinate loss function and classification loss function in the Helicobacter pylori identification neural network model are as follows: Where: L loc Let L be the coordinate loss function. s.c. For the classification loss function, N loc This represents the number of elements in the labeled sub-image. The rectangular box marks the location of Helicobacter pylori. The coordinate information of the i-th rectangle output by the neural network model for identifying Helicobacter pylori is provided. For manually marked rectangular boxes, identify the coordinate information, F loc (·) represents the GIoU loss function, N a N represents the width of the labeled sub-image. o ρ represents the height of the labeled sub-image. mn The value represents the probability that the Helicobacter pylori marker indicated by the rectangle is positive, where i is the coordinate index of the rectangle marker, m is the width index of the labeled sub-image, and n is the height index of the labeled sub-image.
9. The method for identifying Helicobacter pylori based on artificial intelligence fluorescence imaging according to claim 1, characterized in that, Based on the coordinates of the rectangular frame and the coordinates of the predetermined orientation of the sub-image, the coordinates of the rectangular frame are converted into coordinate information in the Helicobacter pylori fluorescent staining image, specifically as follows: Among them: (G) X G Y (S) represents the global coordinates of the target Helicobacter pylori in the original image. X ,S Y (L) represents the coordinates of the origin of the sub-image in the original image. X ,L Y W represents the local coordinates of the target Helicobacter pylori in the subgraph. s H is the pixel width of the sub-image. s This represents the pixel height of the sub-image.
10. A device for identifying Helicobacter pylori based on artificial intelligence fluorescence imaging, characterized in that, The method for identifying Helicobacter pylori based on artificial intelligence fluorescence imaging as described in any one of claims 1-9 was used.
