Gastrointestinal tract real-time imaging and positioning method based on artificial intelligence

Abstract

The invention relates to the technical field of medical imaging, and discloses an artificial intelligence-based gastrointestinal tract real-time imaging and positioning method, which comprises the following steps of: acquiring a gastrointestinal tract image of a patient by adopting medical imaging equipment, and preprocessing the gastrointestinal tract image; the preprocessed gastrointestinal tract image is transmitted to a trained AI deep learning model for real-time analysis, a lesion area is obtained, the lesion area is subjected to highlight marking, and a diagnosis result is generated; packaging the lesion area and the diagnosis result into basic prediction information, and inputting the basic prediction information into the LIME; the LIME performs interpretability analysis on the basic prediction information, generates interpretation of the basic prediction information and generates an interpretation graph; a display screen is adopted to superpose information processed by the AR system on the three-dimensional gastrointestinal tract model, so that a lesion area is positioned, and an interpretation graph is displayed. According to the method, the problem that the output result lacks interpretability due to the black box characteristic of the AI model in the prior art can be solved.

Description

Technical Field

[0001] The present invention relates to the field of medical imaging technology, and in particular to a real-time gastrointestinal tract imaging and positioning method based on artificial intelligence. Background Art

[0002] Gastrointestinal diseases are common types of diseases worldwide. With the continuous development of medical imaging technology, the acquisition and analysis of gastrointestinal images have become an important means of diagnosing gastrointestinal diseases. However, traditional gastrointestinal image analysis methods rely on manual observation and empirical judgment, and have certain limitations, especially in the accurate positioning of lesion areas, the identification of early lesions, and the accuracy of diagnosis. With the development of computer vision, deep learning, and artificial intelligence (AI) technology, AI-based medical image analysis methods have begun to play an important role in medical imaging diagnosis. Especially in the field of gastrointestinal imaging, AI can automatically identify various gastrointestinal lesions (such as tumors, inflammation, ulcers, etc.) and accurately locate them by analyzing a large amount of medical imaging data.

[0003] However, although artificial intelligence technology has made significant progress in the field of medical imaging, it still faces some technical challenges in gastrointestinal imaging and positioning. The "black box" nature of the AI ​​model makes its output results lack interpretability, which creates obstacles to clinicians' use and trust.

[0004] Therefore, researching and designing a technology that combines deep learning, local interpretable models and augmented reality (AR) to provide a more accurate, transparent and easy-to-operate gastrointestinal disease diagnosis solution has become a technical problem that needs to be solved urgently. Summary of the invention

[0005] In view of this, the present invention proposes a real-time gastrointestinal imaging and positioning method based on artificial intelligence, aiming to solve the problem that the "black box" characteristics of the AI ​​model in the current technology make its output results lack interpretability.

[0006] The present invention proposes a gastrointestinal tract real-time imaging and positioning method based on artificial intelligence, comprising:

[0007] Using medical imaging equipment to collect gastrointestinal images of patients, and preprocessing the gastrointestinal images to obtain preprocessed gastrointestinal images;

[0008] The preprocessed gastrointestinal image is transmitted to the trained AI deep learning model for real-time analysis to obtain the lesion area in the preprocessed gastrointestinal image, and the lesion area is highlighted in the preprocessed gastrointestinal image to generate a diagnosis result;

[0009] Packing the lesion area and the diagnosis result into basic prediction information, and inputting the basic prediction information into a local interpretable model-an opaque model;

[0010] The local interpretable model-opaque model performs interpretability analysis on the basic prediction information, generates an explanation of the basic prediction information, and generates an explanation graph;

[0011] The preprocessed gastrointestinal image and lesion area are 3D reconstructed to obtain a three-dimensional gastrointestinal model and a three-dimensional lesion area model, the three-dimensional lesion area model and the interpretation graphics are transmitted to the AR system to obtain AR lesion information, and the AR lesion information is superimposed on the three-dimensional gastrointestinal model using a display screen so that the three-dimensional lesion area model is positioned in the three-dimensional gastrointestinal model and the interpretation graphics are displayed.

[0012] Furthermore, the medical imaging equipment includes endoscope, ultrasound, CT, and MRI.

[0013] Furthermore, the preprocessing of the gastrointestinal tract image specifically includes: denoising, enhancing and contrast adjusting the gastrointestinal tract image.

[0014] Furthermore, the specific content of the trained AI deep learning model is:

[0015] Extracting multiple gastrointestinal case images from a historical database, manually annotating lesion areas in the multiple gastrointestinal case image datasets, and using the multiple gastrointestinal case images to form a gastrointestinal case image dataset;

[0016] The gastrointestinal case dataset is divided into a training set and a validation set in a ratio of 7:3. The training set is placed in the AI ​​deep learning model and trained using the deep learning algorithm to obtain the first AI deep learning model. The validation set is substituted into the first AI deep learning model for verification to obtain a trained AI deep learning model.

[0017] Furthermore, the specific steps of denoising the gastrointestinal image include:

[0018] The gastrointestinal image is denoised to obtain the denoised gastrointestinal image, which is expressed as:

[0019]

[0020] Among them, G ( i,j ) is a two-dimensional Gaussian function, I ( x,y ) For gastrointestinal images at coordinates ( x,y ) The pixel value at I ( x+i,y+j) For gastrointestinal images at coordinates ( x+i,y+j ) The pixel value at ( i,j ) is the filter index coordinate, I denoised( x,y ) is the pixel value of the gastrointestinal image after denoising.

[0021] Furthermore, the specific steps of enhancing the gastrointestinal tract image include:

[0022] The gastrointestinal image is enhanced to obtain the enhanced gastrointestinal image, which is expressed as:

[0023]

[0024] Among them, α1 represents the enhancement coefficient, is the Laplace operator of the image, that is, the second-order derivative of the image, I enhanced( x,y ) is the pixel value of the enhanced gastrointestinal image.

[0025] Furthermore, the specific steps of adjusting the contrast of the gastrointestinal tract image include:

[0026] The contrast of the gastrointestinal image is adjusted to obtain a contrast-adjusted gastrointestinal image, which is expressed as:

[0027] I contrast ( x,y ) =α2·I ( x,y ) -I mean +I mean ;

[0028] Among them, α2 is the contrast adjustment coefficient, I mean is the average value of all pixels in the gastrointestinal image, I contrast( x,y ) is the pixel value of the gastrointestinal image after contrast adjustment;

[0029] The denoised gastrointestinal image pixel values, the enhanced gastrointestinal image pixel values ​​and the contrast-adjusted gastrointestinal image pixel values ​​together constitute a preprocessed gastrointestinal image.

[0030] Furthermore, the lesion area and the diagnosis result are packaged into basic prediction information, and the basic prediction information is input into the local interpretable model-opaque model, specifically comprising: packaging the lesion area and the diagnosis result into a structured data packet in JSON data format and transmitting the data packet to the local interpretable model-opaque model;

[0031] The lesion area includes pixel coordinates and size of the lesion area.

[0032] Furthermore, the local interpretable model-opaque model performs an interpretability analysis on the basic prediction information, generates an explanation of the basic prediction information, and generates an explanation graph. The specific content is: the local interpretable model-opaque model locally perturbs the lesion area in the basic prediction information to generate a local perturbation data set, the local interpretable model-opaque model is trained with the local perturbation data set to obtain a local interpretable model, the local interpretable model-opaque model interprets the results in the basic prediction information, generates an explanation of the basic prediction information, and generates an explanation graph.

[0033] Furthermore, the specific content of using a display screen to superimpose AR lesion information on the three-dimensional gastrointestinal model so that the three-dimensional lesion area model is positioned in the three-dimensional gastrointestinal model and an explanatory graphic is displayed is: the AR system aligns the three-dimensional lesion area model with the three-dimensional gastrointestinal model using an image tracking method, displays them on the display screen after alignment, locates the lesion area, and synchronously displays the explanatory graphic.

[0034] Compared with the prior art, the beneficial effects of the present invention are as follows: the present invention is a gastrointestinal real-time imaging and positioning method based on artificial intelligence, which can perform efficient automatic analysis of gastrointestinal images, accurately identify and mark lesion areas by using a trained deep learning AI model. By highlighting the lesion area in the pre-processed gastrointestinal image, the doctor can obtain the specific location and morphology of the lesion in a short time, and improve the accuracy of lesion detection, especially in the case of complex gastrointestinal tract, small or hidden lesions; the present invention uses a local interpretable model (such as LI ME, SHAP, etc.) to analyze the basic prediction information, and provide a clear explanation for the doctor to help understand how the AI ​​model makes a lesion diagnosis. Specifically, through the training and analysis of the local perturbation data set, the model can reveal the impact of each feature on the final diagnosis, help doctors identify which image features have a decisive impact on the diagnosis results, and thus improve the doctor's trust in AI judgments; combined with augmented reality (AR) technology, the system can align the three-dimensional reconstructed model of the lesion area with the three-dimensional image model of the gastrointestinal tract, and superimpose the information of the lesion area and the diagnosis results in real time. This intuitive AR display method can not only accurately locate the lesion area, but also simultaneously display the diagnostic basis of the AI ​​model, enhance the doctor's confidence in decision-making, and enable doctors to make better decisions in a dynamic environment by synchronously displaying explanatory graphics; through real-time automated analysis and positioning, the system greatly reduces the time of manual analysis, helps doctors obtain diagnostic information faster, and reduces the possibility of human misdiagnosis. Especially in a busy clinical environment, this automated method can significantly improve the doctor's work efficiency, help them focus on more complex cases, and reduce the risk of missed or misdiagnosed cases; through AR technology, doctors can see the lesion location and explanatory graphics more intuitively, which improves the transparency and participation of patients when receiving diagnosis. Patients no longer rely solely on the doctor's verbal explanation, but can also intuitively understand the condition through visual graphics. This interactive method helps to improve patients' understanding and compliance with treatment plans, thereby promoting the improvement of treatment effects; AI deep learning models can be trained by continuously collecting new case data. As the amount of data increases, the accuracy and robustness of the model will continue to improve, and it can better adapt to the individual differences and disease manifestations of different patients, thereby providing more personalized diagnostic results. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present invention. Moreover, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:

[0036] Figure 1The present invention is a flowchart of a method for real-time gastrointestinal imaging and positioning based on artificial intelligence. DETAILED DESCRIPTION

[0037] Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided in order to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art. It should be noted that, in the absence of conflict, the embodiments of the present invention and the features described in the embodiments can be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0038] See also Figure 1 As shown, an embodiment of the present invention provides a gastrointestinal tract real-time imaging and positioning method based on artificial intelligence, comprising:

[0039] S1: using a medical imaging device to collect a gastrointestinal image of a patient, and preprocessing the gastrointestinal image to obtain a preprocessed gastrointestinal image;

[0040] S2: The preprocessed gastrointestinal image is transmitted to the trained AI deep learning model for real-time analysis to obtain the lesion area in the preprocessed gastrointestinal image, and the lesion area is highlighted in the preprocessed gastrointestinal image to generate a diagnosis result;

[0041] S3: Pack the lesion area and the diagnosis result into basic prediction information, and input the basic prediction information into the local interpretable model - opaque model;

[0042] S4: Local interpretable model-opaque model performs interpretability analysis on the basic prediction information, generates explanations for the basic prediction information, and generates explanation graphics;

[0043] S5: Perform 3D reconstruction on the preprocessed gastrointestinal image and lesion area to obtain a three-dimensional gastrointestinal model and a three-dimensional lesion area model, transmit the three-dimensional lesion area model and the interpretation graphics to the AR system to obtain AR lesion information, and use a display screen to superimpose the AR lesion information on the three-dimensional gastrointestinal model so that the three-dimensional lesion area model is positioned in the three-dimensional gastrointestinal model and the interpretation graphics are displayed.

[0044] Furthermore, medical imaging equipment includes endoscopes, ultrasound, CT, and MRI.

[0045] Furthermore, the gastrointestinal image is preprocessed by: denoising, enhancing and contrast adjusting the gastrointestinal image.

[0046] Specifically, the convolution kernel of the Gaussian function is used to smooth the gastrointestinal image and reduce the random noise in the gastrointestinal image. Removing noise helps eliminate the adverse effects on the AI ​​deep learning model, especially when the lesion area is small or the contrast with the background tissue is low. Denoising can improve the recognition ability of the model and reduce the occurrence of false positives and false negatives; the second-order derivative of the gastrointestinal image is used to detect the edges in the gastrointestinal image and enhance the details of the gastrointestinal image, especially in areas with poor edge clarity. Through the enhancement processing, the boundaries of the lesion area are clearer, which is convenient for clinicians or AI systems to accurately locate and identify. At the same time, it can enhance the contrast between the lesion area and the surrounding healthy tissue, making the lesion easier to detect, especially in the case of small or early lesions; the contrast of the image is adjusted by a linear function, so that the brightness range of the image is expanded and the details of the lesion area are more obvious. By adjusting the contrast of the image, the visualization effect of the image can be improved, making the subtle anatomical structure and lesion area clearer, especially in low light and complex backgrounds, which can significantly improve the readability of the image.

[0047] Furthermore, the specific contents of the trained AI deep learning model are as follows:

[0048] Extracting multiple gastrointestinal case images from a historical database, manually annotating lesion areas in the multiple gastrointestinal case image datasets, and using the multiple gastrointestinal case images to form a gastrointestinal case image dataset;

[0049] The gastrointestinal case dataset is divided into a training set and a validation set in a ratio of 7:3. The training set is placed in the AI ​​deep learning model and trained using the deep learning algorithm to obtain the first AI deep learning model. The validation set is substituted into the first AI deep learning model for verification to obtain a trained AI deep learning model.

[0050] Specifically, case data containing gastrointestinal images are extracted from the historical database. For each gastrointestinal image, the lesion area is manually annotated, and the specific type and location of the lesion are recorded. These annotated data are usually stored in rectangular boxes, outlines or other forms, and correspond one-to-one with the original images. The annotated images and corresponding label information (lesion area, type) are combined into a data set for subsequent training. The gastrointestinal case data set is divided into a training set and a validation set in a ratio of 7:3. The training set is used to train the AI ​​deep learning model using a convolutional neural network algorithm. The validation set is input into the trained model for prediction. The performance of the model is evaluated by calculating the accuracy, recall, F1-score and other indicators of the model on the validation set.

[0051] It should be noted that by using a variety of case images, the AI ​​deep learning model can learn various lesion types and different gastrointestinal imaging features, improve its ability to recognize unseen cases, and manually label the lesion areas to ensure that the training data of the AI ​​deep learning model is of high quality and meets clinical requirements, thus avoiding the deviations that may be caused by automatic labeling.

[0052] Furthermore, the specific steps of denoising the gastrointestinal image include:

[0053] The gastrointestinal image is denoised to obtain the denoised gastrointestinal image, which is expressed as:

[0054]

[0055] Among them, G ( i,j ) is a two-dimensional Gaussian function, I ( x,y ) For gastrointestinal images at coordinates ( x,y ) The pixel value at I ( x+i,y+j ) For gastrointestinal images at coordinates ( x+i,y+j ) The pixel value at ( i,j ) is the filter index coordinate, which is used to define the size of the filter. k determines the radius of the filter. denoised( x,y ) is the pixel value of the gastrointestinal image after denoising.

[0056] Specifically, G. ( i,j ) is the Gaussian filter kernel, that is, a two-dimensional Gaussian function, expressed as:

[0057]

[0058] Among them, σ is the standard deviation of the Gaussian distribution, which controls the strength of denoising.

[0059] Furthermore, the specific steps of enhancing the gastrointestinal image include:

[0060] The gastrointestinal image is enhanced to obtain the enhanced gastrointestinal image, which is expressed as:

[0061]

[0062] Among them, α1 represents the enhancement coefficient, is the Laplace operator of the image, that is, the second-order derivative of the image, I enhanced( x,y )is the pixel value of the enhanced gastrointestinal image.

[0063] Specifically, is the Laplace operator of the image, that is, the second-order derivative of the image, expressed as:

[0064]

[0065] Among them, L ( i,j ) is the Laplace filter kernel, expressed as:

[0066]

[0067] Furthermore, the specific steps of contrast adjustment for the gastrointestinal tract image include:

[0068] The contrast of the gastrointestinal image is adjusted to obtain a contrast-adjusted gastrointestinal image, which is expressed as:

[0069] I contrast ( x,y ) =α2·I ( x,y ) -I mean +I mean ;

[0070] Among them, α2 is the contrast adjustment coefficient, I mean is the average value of all pixels in the gastrointestinal image, I contrast( x,y ) is the pixel value of the gastrointestinal image after contrast adjustment;

[0071] The denoised gastrointestinal image pixel values, the enhanced gastrointestinal image pixel values ​​and the contrast-adjusted gastrointestinal image pixel values ​​together constitute the preprocessed gastrointestinal image.

[0072] Specifically, I mean is the average value of all pixels in the gastrointestinal image, and the calculation formula is:

[0073]

[0074] Wherein, W is the width of the gastrointestinal tract image, H is the height of the gastrointestinal tract image, and N is the total number of pixels.

[0075] Furthermore, the lesion area and the diagnosis result are packaged into basic prediction information, and the basic prediction information is input into the local interpretable model-opaque model. Specifically, the lesion area and the diagnosis result are packaged into a structured data packet in JSON data format and transmitted to the local interpretable model-opaque model;

[0076] The lesion area includes the pixel coordinates and size of the lesion area.

[0077] Specifically, the gastrointestinal images are first analyzed through an AI deep learning model to identify and locate lesion areas (such as tumors, ulcers, polyps, etc.). These lesion areas are usually specific parts of the image and contain information that is crucial for disease diagnosis. In order to facilitate subsequent analysis and model interpretation, the specific location and size of the lesion area need to be transmitted in a structured format. At this time, the JSON (JavaScript Object Notation) format can efficiently and easily store and transmit data in a human-readable manner.

[0078] It should be noted that by structuring the relevant information of the lesion area (such as pixel coordinates and size) into JSON format, it not only facilitates the subsequent analysis and use of data, but also improves the efficiency and accuracy of data transmission. Transmitting lesion area data in JSON format can flexibly interact between different systems and algorithms, avoiding problems that may be caused by inconsistent formats. After receiving JSON data, the local interpretable model (such as LI ME) can perform further analysis based on the specific characteristics of the lesion area. The specific information of the lesion area is transmitted in a structured data format, which helps the interpretable model understand the key areas that the deep learning model focuses on during image analysis, thereby improving the transparency of the results. This allows the final diagnostic results to be better understood and verified by doctors.

[0079] Furthermore, the local interpretable model-opaque model performs interpretability analysis on the basic prediction information, generates an explanation of the basic prediction information, and generates an explanation graph. The specific content is: the local interpretable model-opaque model locally perturbs the lesion area in the basic prediction information to generate a local perturbation data set, the local interpretable model-opaque model is trained with the local perturbation data set to obtain a local interpretable model, the local interpretable model-opaque model interprets the results in the basic prediction information, generates an explanation of the basic prediction information, and generates an explanation graph.

[0080] Specifically, first, based on the output of the AI ​​deep learning model, the lesion area (such as a tumor, ulcer or polyp) in the gastrointestinal image is determined, and the lesion area is locally perturbed, that is, the pixel value of the lesion area is slightly adjusted. For example, the pixel value of the lesion area is changed, the lesion area is rotated, the size of the lesion area is adjusted, or the color of the lesion area is changed. Each change will generate a new "perturbed image". These fine-tuned images are combined with the original image to form a local perturbation data set. Each item in the data set has a corresponding label (such as whether it is still predicted to be a lesion, or the classification of the lesion). Secondly, based on the local perturbation data set, a simple interpretable model (such as linear regression, decision tree, etc.) is used to train the data to obtain a local interpretable model. The local interpretable model will learn which features have a greater impact on the prediction results by fitting each data item in the local perturbation data set. For example, it may be found that "color changes in the lesion area" have a greater impact on the prediction results than "size changes in the lesion area". Through the training process, the local interpretable model will generate a set of weights or coefficients, which represent the degree of influence of each feature (such as different attributes of the lesion area) on the prediction results. Finally, the local interpretable model is used to analyze the basic prediction information, especially the characteristics of the lesion area (such as size, shape, location, etc.), and analyze how these features affect the model prediction. According to the output of the local interpretable model, an explanation graph is generated. The explanation graph can include heat maps: showing the contribution of different parts of the lesion area to the model prediction results. The darker the color, the greater the influence of the area on the prediction results. Annotation images: superimpose different annotations (such as "key areas" or "key features") and descriptions on the gastrointestinal images to highlight the impact of the lesion area and help doctors understand the reasoning process of the model.

[0081] It should be noted that through this local perturbation, the output changes of the model under different perturbations can be observed, so as to understand the sensitivity of the model in a specific area. The perturbed data set can help identify which features (such as the location, shape or size of the lesion area) are important for the prediction results, and which features have no significant impact on the model decision. The deep learning model is essentially a "black box" model. Through the training of the local interpretable model, the complex prediction process can be transformed into a more understandable form, so that experts can understand how the model makes predictions through different features of the lesion area. The trained local interpretable model can provide clinicians with more details, explain the decision-making basis of the model on the lesion area, help doctors understand why certain features are considered important, and then support medical decision-making. By generating explanatory graphics, the prediction logic of the AI ​​model can be intuitively presented, making it easier for doctors and other medical professionals to understand the decision-making process of the deep learning model. Explanatory graphics help transform the "black box" model into a form that is easy to understand and verify.

[0082] Furthermore, a display screen is used to superimpose AR lesion information on the three-dimensional gastrointestinal model so that the three-dimensional lesion area model is positioned in the three-dimensional gastrointestinal model and an explanatory graphic is displayed. The specific content is: the AR system aligns the three-dimensional lesion area model with the three-dimensional gastrointestinal model using an image tracking method, and displays them on the display screen after alignment, locates the lesion area, and simultaneously displays the explanatory graphic.

[0083] Specifically, first, the AR system captures the real-time image of the patient's gastrointestinal tract through a camera and extracts features from the image. Common features include corners, edges, textures, etc.; secondly, the AR system calculates the similarity between the real-time image and the feature points in the 3D model, determines the corresponding 3D model position in the image through a matching algorithm, and estimates the image's pose (i.e., position and posture in 3D space) by calculating the transformation relationship between the features extracted from the image and the 3D model. This estimation can be achieved through methods such as PnP (Perspective-n-Poin) algorithm and SLAM (Simultaneous Localization and Mapping). Finally, based on the estimated pose information, the AR system aligns the 3D lesion area model and the 3D gastrointestinal tract model to the real-time image. In this way, the virtual model can be accurately superimposed on the gastrointestinal tract area in the real image.

[0084] It should be noted that image tracking technology can ensure that the three-dimensional virtual lesion area model and gastrointestinal model are completely aligned with the actual gastrointestinal image position on the display screen, so that the lesion area can be accurately located on the display screen, helping doctors see the specific location of the lesion area. Image tracking can process and update images in real time. When the gastrointestinal image or patient position changes, the AR system will immediately update the position of the model to ensure that the virtual model is always synchronized with the actual image to avoid alignment errors.

[0085] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A gastrointestinal tract real-time imaging and positioning method based on artificial intelligence, characterized in that: include: Using medical imaging equipment to collect gastrointestinal images of patients, and preprocessing the gastrointestinal images to obtain preprocessed gastrointestinal images; The preprocessed gastrointestinal image is transmitted to the trained AI deep learning model for real-time analysis to obtain the lesion area in the preprocessed gastrointestinal image, and the lesion area is highlighted in the preprocessed gastrointestinal image to generate a diagnosis result; Packing the lesion area and the diagnosis result into basic prediction information, and inputting the basic prediction information into a local interpretable model-an opaque model; The local interpretable model-opaque model performs interpretability analysis on the basic prediction information, generates an explanation of the basic prediction information, and generates an explanation graph; The preprocessed gastrointestinal image and lesion area are 3D reconstructed to obtain a three-dimensional gastrointestinal model and a three-dimensional lesion area model, the three-dimensional lesion area model and the interpretation graphics are transmitted to the AR system to obtain AR lesion information, and the AR lesion information is superimposed on the three-dimensional gastrointestinal model using a display screen so that the three-dimensional lesion area model is positioned in the three-dimensional gastrointestinal model and the interpretation graphics are displayed.

2. The method for real-time gastrointestinal tract imaging and positioning based on artificial intelligence according to claim 1, characterized in that: The medical imaging equipment includes endoscope, ultrasound, CT, MRI.

3. The method for real-time gastrointestinal tract imaging and positioning based on artificial intelligence according to claim 2, characterized in that: The specific content of the preprocessing of the gastrointestinal tract image is: denoising, enhancing and contrast adjusting the gastrointestinal tract image.

4. The method for real-time gastrointestinal tract imaging and positioning based on artificial intelligence according to claim 3, characterized in that: The specific content of the trained AI deep learning model is: Extracting multiple gastrointestinal case images from a historical database, manually annotating lesion areas in the multiple gastrointestinal case image datasets, and using the multiple gastrointestinal case images to form a gastrointestinal case image dataset; The gastrointestinal case dataset is divided into a training set and a validation set in a ratio of 7:

3. The training set is placed in the AI ​​deep learning model and trained using the deep learning algorithm to obtain the first AI deep learning model. The validation set is substituted into the first AI deep learning model for verification to obtain a trained AI deep learning model.

5. The method for real-time gastrointestinal tract imaging and positioning based on artificial intelligence according to claim 4, characterized in that: The specific steps of denoising the gastrointestinal image include: The gastrointestinal image is denoised to obtain the denoised gastrointestinal image, which is expressed as: Among them, G ( i,j ) is a two-dimensional Gaussian function, I ( x,y ) For gastrointestinal images at coordinates ( x,y ) The pixel value at I ( x+i,y+j ) For gastrointestinal images at coordinates ( x+i,y+j ) The pixel value at ( i,j ) is the filter index coordinate, I denoised( x,y ) is the pixel value of the gastrointestinal image after denoising.

6. The method for real-time gastrointestinal tract imaging and positioning based on artificial intelligence according to claim 5, characterized in that: The specific steps of enhancing the gastrointestinal tract image include: The gastrointestinal image is enhanced to obtain the enhanced gastrointestinal image, which is expressed as: Among them, α1 represents the enhancement coefficient, is the Laplace operator of the image, that is, the second-order derivative of the image, I enhanced( x,y ) is the pixel value of the enhanced gastrointestinal image.

7. The method for real-time gastrointestinal tract imaging and positioning based on artificial intelligence according to claim 6, characterized in that: The specific steps of adjusting the contrast of the gastrointestinal tract image include: The contrast of the gastrointestinal image is adjusted to obtain a contrast-adjusted gastrointestinal image, which is expressed as: I contrast ( x,y ) =α2·I ( x,y ) -I mean +I mean ; Among them, α2 is the contrast adjustment coefficient, I mean is the average value of all pixels in the gastrointestinal image, I contrast( x,y ) is the pixel value of the gastrointestinal image after contrast adjustment; The denoised gastrointestinal image pixel values, the enhanced gastrointestinal image pixel values ​​and the contrast-adjusted gastrointestinal image pixel values ​​together constitute a preprocessed gastrointestinal image.

8. The method for real-time gastrointestinal tract imaging and positioning based on artificial intelligence according to claim 7, characterized in that: The step of packaging the lesion area and the diagnosis result into basic prediction information and inputting the basic prediction information into the local interpretable model-opaque model specifically comprises: packaging the lesion area and the diagnosis result into a structured data packet in the JSON data format and transmitting the data packet to the local interpretable model-opaque model; The lesion area includes pixel coordinates and size of the lesion area.

9. The method for real-time gastrointestinal tract imaging and positioning based on artificial intelligence according to claim 8, characterized in that: The local interpretable model-opaque model performs interpretability analysis on the basic prediction information, generates an explanation of the basic prediction information, and generates an explanation graph. The specific content is: the local interpretable model-opaque model performs local perturbations on the lesion area in the basic prediction information to generate a local perturbation data set, the local interpretable model-opaque model is trained with the local perturbation data set to obtain a local interpretable model, the local interpretable model-opaque model interprets the results in the basic prediction information, generates an explanation of the basic prediction information, and generates an explanation graph.

10. The method for real-time gastrointestinal tract imaging and positioning based on artificial intelligence according to claim 9, characterized in that: The specific content of using a display screen to superimpose AR lesion information on the three-dimensional gastrointestinal model so that the three-dimensional lesion area model is positioned in the three-dimensional gastrointestinal model and an explanatory graphic is displayed is: the AR system aligns the three-dimensional lesion area model with the three-dimensional gastrointestinal model using an image tracking method, displays them on the display screen after alignment, locates the lesion area, and synchronously displays the explanatory graphic.

Images