Gastroendoscope real-time auxiliary system and method based on artificial intelligence

Abstract

The invention discloses a digestive endoscopy real-time auxiliary system and method based on artificial intelligence, relates to the technical field of image processing, solves the problem that an original lesion area calibration mode is not accurate and obvious, and provides a real-time auxiliary system and method for a digestive endoscopy by accurately determining the outer contour and the inner contour of a contour area and then selecting bisectors and feature midpoints. The contour area is calibrated again, and the contour features of the focus area can be clearly highlighted; therefore, medical staff can judge the boundary and the range of a focus more accurately, the recognition capability of tiny lesions can be improved, the possibility of misdiagnosis and missed diagnosis is reduced, a more reliable basis is provided for clinical diagnosis, finally, the overall quality of digestive endoscopy is improved, and patients are assisted to obtain more timely and accurate medical diagnosis and treatment.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to a real-time digestive endoscopy assistance system and method based on artificial intelligence. Background Art

[0002] In modern medicine, digestive endoscopy plays an indispensable role as an important tool for diagnosing digestive system diseases. Through digestive endoscopy, doctors can directly observe the internal conditions of the digestive tract, providing a key basis for accurate disease diagnosis. However, in practice, traditional digestive endoscopy faces many challenges.

[0003] The microscopic examination process generates a large number of microscopic images, which are huge in number and contain complex information. Medical staff rely on manual screening and analysis of these images one by one, which is not only time-consuming and labor-intensive, but also extremely inefficient. It is also easy to miss key information due to factors such as visual fatigue, affecting the accuracy of diagnosis. In traditional image analysis, there is a lack of effective automated means to classify images, making it difficult to quickly extract image sets with similar features from massive images, increasing the difficulty and time cost of diagnosis.

[0004] When determining representative images, traditional methods often lack scientific and quantitative standards, making it difficult to accurately select images that best reflect the characteristics of the lesion, greatly reducing the targeted nature of the diagnosis. Traditional methods are also relatively rough in analyzing the contours of the lesion area in the image, and are unable to accurately define the boundaries and scope of the lesion. In particular, the recognition ability is limited for subtle lesions, which can easily lead to misdiagnosis and missed diagnosis, seriously affecting the effectiveness of gastrointestinal endoscopy in disease diagnosis. Summary of the Invention

[0005] In response to the shortcomings of the existing technology, the present invention provides a real-time auxiliary system and method for digestive endoscopy based on artificial intelligence, which solves the problem that the original method of calibrating the lesion area is not accurate and obvious.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: a real-time assisted digestive endoscopy method based on artificial intelligence, comprising the following steps:

[0007] Step 1: Process the microscopic images generated during the digestive endoscopy in real time. Use the Sobel algorithm to identify the contour area in each frame of the microscopic image. Based on the overall contour features of the contour area, classify several frames of images and identify several groups of feature image sets. The specific method is as follows:

[0008] Grayscale processing is performed on each frame of the microscopic image to confirm the grayscale image associated with the corresponding microscopic image;

[0009] The Sobel algorithm is used to confirm the vertical gradient and vertical gradient associated with different pixels in the grayscale image, and the vertical gradient associated with the corresponding pixel is calibrated as C i , and the associated vertical gradient is calibrated as Z i , where i represents different pixels;

[0010] use Confirm the comprehensive gradient ZH i , will satisfy: ZH i Pixels with Y1 > are calibrated as gradient pixels. Otherwise, no calibration is performed and Y1 is the preset value.

[0011] Based on a number of groups of gradient pixel points identified in the grayscale image, continuous gradient pixel points are connected to identify contour lines, and the middle area contained in the contour lines is marked as the contour area;

[0012] Confirm the center point of the contour area associated with the grayscale image, and use the two-dimensional coordinate system to confirm the two-dimensional coordinates associated with different contour points in the contour line corresponding to the contour area. Then, perform mean processing on several groups of two-dimensional coordinates to confirm the mean coordinates. Based on the confirmed mean coordinates, perform center point calibration in the contour area.

[0013] If there is only one set of contour areas in the grayscale image, the center point of the contour area is recorded as the feature point of the grayscale image;

[0014] If there are two sets of contour areas in the grayscale image, the midpoint of the line connecting the center points of the two sets of contour areas is recorded as the feature point of the grayscale image;

[0015] If there are more than two groups of contour areas in the grayscale image, connect the center points of the adjacent contour areas to determine a group of polygons, and record the center points of these polygons as the feature points of the grayscale image;

[0016] Confirm the position of feature points of grayscale images of different frames that appear in sequence. If the distance between the feature points of adjacent grayscale images is lower than the threshold, they are classified into the same type of grayscale images. If it is not lower than the threshold, the grayscale images of the same type are not calibrated. Multiple frames of different grayscale images belonging to the same type of grayscale images are divided into the same set of feature images.

[0017] Step 2: for the confirmed feature image set, different grayscale images are used to confirm the area of ​​the contour region, and the grayscale image with the largest contour region area is selected as the feature image of the corresponding feature image set;

[0018] Step 3: Refine the features of the feature images confirmed in different feature image sets, confirm the contour line areas of different contour areas in the feature images, and select feature midpoints from the contour line areas based on the pixel value change characteristics of the pixels inside the contour line areas. Based on the locations of several feature midpoints, refine the features of the contour areas. The specific method is as follows:

[0019] Based on the contour line area associated with the contour area, connect the outermost gradient pixel points from the center point of the contour area to confirm the outer contour: based on the location of the center point of the area, determine the gradient pixel point farthest from the current position, and draw a circle with the radius of the center point of the area and this gradient pixel point. Then, sequentially determine the gradient pixel points closest to the circumference in the contour line area and record them as outer contour points. Connect several outer contour points to confirm the outer contour;

[0020] Then connect the innermost gradient pixel points from the center point of the contour area to confirm the internal contour: determine the gradient pixel point closest to the current position based on the location of the center point of the area, and draw a circle with the radius of the center point of the area and the gradient pixel point. Then, sequentially determine the gradient pixel points closest to the circumference of the circle in the contour connection area and record them as internal contour points. Connect several internal contour points to confirm the internal contour;

[0021] Based on the confirmed outer contour and inner contour, contour points are sequentially selected and connected in the outer contour and the inner contour to determine a number of bisectors, wherein the bisectors divide the contour connection area into n areas of equal area, where n is a preset value;

[0022] Identify the pixels associated with each set of bisectors and mark them as the pending pixels of this bisector. Then select the equilibrium point among several pending pixels and record it as the feature midpoint.

[0023] Confirm the feature midpoints of several equally divided lines in turn, connect the confirmed feature midpoints, confirm the feature connection line, and then use this feature connection line as the contour connection line of this contour area, and recalibrate the contour area to complete the feature refinement process of this contour area;

[0024] The feature images confirmed in different feature image sets are refined in turn and displayed through specific display terminals.

[0025] The specific method of selecting a balance point from among several undetermined pixel points and recording it as the feature midpoint is as follows:

[0026] Sort the pixel values ​​of several pending pixels from one end point of the bisector to the other end point, and confirm whether the sorted pixel values ​​show a gradually increasing or decreasing state:

[0027] If yes, then the maximum value is selected from the sorted multiple pixel values, and the undetermined pixel point corresponding to the maximum value is marked as the feature midpoint;

[0028] If not, then select a point from the middle pending pixel points included in the bisector, record this point and several pending pixel points between the endpoint on one side as one side pixel point set, record this point and several pending pixel points between the endpoint on the other side as the other side pixel point set, calculate the difference between the pixel value of this point and the pixel value of the pixel point set on one side in turn, and determine the sum of the differences and record it as one side feature, then calculate the difference between the pixel value of the point and the pixel value of the pixel point set on the other side in turn, and determine the sum of the differences and record it as the other side feature, confirm the interval value between the feature on one side and the feature on the other side, and use this interval value as the process feature of the selection process of this point, then select other middle pending pixel points in turn, and determine the process feature of the corresponding selected process, lock the pending pixel point associated with the minimum process feature from several process features and record it as the feature midpoint.

[0029] Preferably, a real-time digestive endoscopy assistance system based on artificial intelligence comprises:

[0030] The grayscale image conversion end performs grayscale processing on the microscopic images generated during the microscopic examination of the digestive endoscope, and confirms the grayscale image associated with the corresponding microscopic image;

[0031] On the feature classification side, the contour area in each frame of the microscopic image is confirmed based on the Sobel algorithm, and based on the overall contour features of the contour area, several frames of images are classified to confirm several groups of feature image sets;

[0032] The feature image calibration end verifies the area of ​​the contour region of the confirmed feature image set using different grayscale images, and selects the grayscale image with the largest contour region area as the feature image of the corresponding feature image set;

[0033] The contour feature refinement end refines the feature images confirmed in different feature image sets, confirms the contour line areas of different contour areas in the feature images, and selects feature midpoints from the contour line areas based on the pixel value change characteristics of the pixel points inside the contour line areas, and refines the features of the contour areas based on the locations of several feature midpoints.

[0034] The present invention provides a real-time digestive endoscopy assistance system and method based on artificial intelligence. Compared with the existing technology, it has the following advantages:

[0035] This method uses the Sobel algorithm to identify contour areas and classify images, which can quickly screen out similar feature images from a large number of microscopic images, greatly improving the efficiency of image processing and laying the foundation for subsequent accurate analysis. Compared with traditional manual screening, it greatly saves medical staff time and energy;

[0036] The image with the largest contour area is used as the feature image, so that each set of feature images can present the most representative features, making it easier for medical staff to quickly focus on key images and improve the pertinence and accuracy of diagnosis;

[0037] By accurately determining the outer and inner contours of the contour area, selecting the bisectors and feature midpoints, and recalibrating the contour area, the contour features of the lesion area can be clearly highlighted. This not only helps medical staff to more accurately judge the boundaries and scope of the lesions, but also improves the ability to identify subtle lesions, reduces the possibility of misdiagnosis and missed diagnosis, provides a more reliable basis for clinical diagnosis, and ultimately improves the overall quality of digestive endoscopy, helping patients receive more timely and accurate medical diagnosis and treatment. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 Schematic diagram of the process of the present invention;

[0039] Figure 2 It is a schematic diagram of the principle framework of the present invention. DETAILED DESCRIPTION

[0040] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only 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 ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0041] First embodiment

[0042] See also Figure 1 , the present application provides a real-time assisted method for digestive endoscopy based on artificial intelligence, comprising the following steps:

[0043] Step 1: Process the microscopic images generated by the digestive endoscope in real time during the microscopic examination process. Confirm the contour area in each frame of the microscopic image based on the Sobel algorithm, and classify several frames of images based on the overall contour features of the contour area to confirm several groups of feature image sets. Specifically, during the microscopic examination process, a corresponding microscopic video will be generated. The video is generated by continuously playing several frames of images. Therefore, the display features associated with each frame of the image have corresponding contour features. From the contour features, image classification can be performed, and a large number of images with similar contour features can be specifically classified to facilitate subsequent feature confirmation of the classified images. The specific method of confirming the feature image set is as follows:

[0044] Grayscale processing is performed on each frame of the microscopic image to confirm the grayscale image associated with the corresponding microscopic image (grayscale processing is to assign different weights to the corresponding RGB values, confirm the grayscale values ​​of the corresponding points, and then make relevant adjustments to confirm the corresponding grayscale image. The sum of the assigned weights is 1 and is a preset value. This is relatively common in the prior art and will not be described in detail here);

[0045] The Sobel algorithm is used to confirm the vertical gradient and vertical gradient associated with different pixels in the grayscale image (in the Sobel algorithm, for the corresponding pixel point, this pixel point is taken as the center position, and the eight pixels corresponding to the surrounding area are extracted to confirm a group of pixel arrays belonging to this pixel point. The pixel values ​​associated with different pixels are then given different weights, and then summed up to obtain the vertical gradient or vertical gradient associated with the corresponding pixel point. The weight range is generally [-2, 2], and each weight is an integer, including 0. Since the method of determining the pixel gradient value is relatively common in the prior art, it will not be described in detail here). The vertical gradient associated with the corresponding pixel point is calibrated as C i , and the associated vertical gradient is calibrated as Z i , where i represents different pixels;

[0046] use Confirm the comprehensive gradient ZH i , will satisfy: ZH i Pixels with Y1 > are calibrated as gradient pixels. Otherwise, no calibration is performed. Y1 is a preset value, and its specific value is determined by the operator based on experience. Since the features of the lesion area in the corresponding image are more obvious, a value of 30 is generally appropriate for Y1.

[0047] Based on several groups of gradient pixel points confirmed in the grayscale image, continuous gradient pixel points are connected to confirm the contour line, and the middle area contained in the contour line is calibrated as the contour area. If there is only a contour line and there is no middle area contained, no calibration is performed.

[0048] Based on the overall contour features of the contour area, the specific method of classifying several frames of images is as follows:

[0049] Confirm the center point of the contour area associated with the grayscale image, and combine the two-dimensional coordinate system to confirm the two-dimensional coordinates associated with different contour points in the contour line corresponding to the contour area. Then, perform mean processing on several groups of two-dimensional coordinates to confirm the mean coordinates. Based on the confirmed mean coordinates, calibrate the center point in the contour area (the location of the mean coordinate point is the location of the center point of this contour area, so the center point of the corresponding contour area can be quickly and effectively confirmed);

[0050] If there is only one set of contour areas in the grayscale image, the center point of the contour area is recorded as the feature point of the grayscale image;

[0051] If there are two sets of contour areas in the grayscale image, the midpoint of the line connecting the center points of the two sets of contour areas is recorded as the feature point of the grayscale image;

[0052] If there are more than two groups of contour areas in the grayscale image, connect the center points of the adjacent contour areas to determine a group of polygons, and record the center points of these polygons as the feature points of the grayscale image;

[0053] The feature point positions of grayscale images of different frames that appear in sequence are confirmed. If the feature point position distance of adjacent frames of grayscale images is lower than the threshold, they are divided into the same type of grayscale images. If it is not lower than the threshold, the grayscale images of the same type are not calibrated, and multiple frames of different grayscale images belonging to the same type of grayscale images are divided into the same set of feature images. Specifically, the threshold is generally 0.5mm, that is, according to the specific imaging process of endoscopic examination, each frame of grayscale image is sorted in sequence, and the feature points inside it are confirmed in sequence from front to back. The proposed grayscale images are: ABCDE, where A is the first frame image and E is the last frame image. The internal feature points of A and the internal feature points of B meet the judgment features, B and C also meet the judgment features, and C and D do not meet the judgment features. Then A, B and C belong to the same type of grayscale images, and then the same type of grayscale images are confirmed in sequence starting from D. Under normal circumstances, multiple frames of images in the same area can be effectively divided into the same category, which is convenient for subsequent feature analysis and confirmation.

[0054] Step 2: For the confirmed feature image set, perform contour area confirmation on different grayscale images, and select the grayscale image with the largest contour area as the feature image of the corresponding feature image set. The specific method for confirming the feature image is as follows:

[0055] For different grayscale images in the feature image set, the contour areas confirmed in different grayscale images are confirmed, and the sum of the contour areas belonging to a single set of grayscale images is determined and calibrated as Mz k , where k represents different grayscale images;

[0056] Then the total area parameter of the corresponding grayscale image is confirmed and calibrated as Zc k ;

[0057] Use: Z k =Mz k ÷Zc k Confirm the area ratio feature Z of the corresponding contour area k , and then from several different area ratio features Z k , select Z k The grayscale image associated with max is used as the feature image of this feature image set;

[0058] Specifically, in each set of different feature images, there is an image with the largest area parameter expression state, which is the corresponding feature image. This type of image has the most obvious feature expression in comprehensive feature expression, which is convenient for subsequent medical staff to make a diagnosis and also facilitates the comprehensive representation of the corresponding feature contour.

[0059] Step 3: Refine the features of the feature images confirmed in different feature image sets, confirm the contour line areas of different contour areas in the feature images, and select feature midpoints from the contour line area based on the pixel value change characteristics of the pixels inside the contour line area, and refine the features of the contour area based on the positions of several feature midpoints. Specifically, the determined contour area is not just a simple determination of the external gradient points. The edge points of the contour area are in a gradual change form when changing, so there should be a large number of gradient points on the edge. For example, a plain is surrounded by mountains, then the pixel gradient inside the plain is a stable feature, and its peak belongs to a gradient change state. The area included in the peak is a gradient pixel point with abnormal gradient change. In order to make the feature performance of the corresponding contour area most obvious, the corresponding contour area feature is refined to lock the most characteristic contour edge line.

[0060] The specific method of performing contour area feature refinement is as follows:

[0061] Based on the contour line area associated with the contour area, connect the outermost gradient pixel points from the center point of the contour area to confirm the outer contour: based on the location of the center point of the area, determine the gradient pixel point farthest from the current position, and draw a circle with the radius of the center point of the area and this gradient pixel point. Then, sequentially determine the gradient pixel points closest to the circumference in the contour line area and record them as outer contour points. Connect several outer contour points to confirm the outer contour;

[0062] Then connect the innermost gradient pixel points from the center point of the contour area to confirm the internal contour: determine the gradient pixel point closest to the current position based on the location of the center point of the area, and draw a circle with the radius of the center point of the area and the gradient pixel point. Then, sequentially determine the gradient pixel points closest to the circumference of the circle in the contour connection area and record them as internal contour points. Connect several internal contour points to confirm the internal contour;

[0063] Based on the confirmed outer contour and inner contour, contour points are selected in sequence within the outer contour and the inner contour to connect and determine a number of bisectors. The bisectors divide the contour connection area into n equal-sized areas, where n is a preset value with a minimum of 100 and a maximum of 500, which is set in advance by relevant personnel.

[0064] Identify the pixels associated with each set of bisectors (for a more intuitive understanding of the technical solution, the corresponding pixels can be understood as a small grid with uniform distribution) and mark them as the pending pixels of this bisector. Then, select the equilibrium point among several pending pixels and record it as the feature midpoint. The specific method for determination is as follows:

[0065] Sort the pixel values ​​of several undetermined pixels from one end point of the bisector to the other end point, and confirm whether the sorted pixel values ​​are gradually increasing or decreasing (that is, the values ​​are gradually increasing or decreasing):

[0066] If yes, then the maximum value is selected from the sorted multiple pixel values, and the undetermined pixel point corresponding to the maximum value is marked as the feature midpoint;

[0067] If not, then select a point between the middle undetermined pixel points included in the bisector (excluding the two end points), record the several undetermined pixel points between this point and the end point on one side as the pixel point set on one side (including the end point on one side), and record the several undetermined pixel points between this point and the end point on the other side as the pixel point set on the other side (including the end point on the other side). Differences are taken between the pixel value of this point and the pixel value of the pixel point set on one side in sequence, and the sum of the differences is determined and recorded as the feature on one side. Then, differences are taken between the pixel value of the point and the pixel value of the pixel point set on the other side in sequence, and the sum of the differences is determined and recorded as the feature on the other side. Confirm the interval value between the feature on one side and the feature on the other side (that is, 1 and -1, then the interval value is 2, and the interval value is the corresponding distance between the two values). Value), use this interval value as the process feature of the process of selecting this point, and then select other intermediate undetermined pixel points in turn, and determine the process feature of the corresponding selected process, lock the undetermined pixel point associated with the minimum process feature from several process features and record it as the feature midpoint, for example: suppose the pixel point set on one side is {T1, T2, T3}, and the pixel point set on the other side is {T5, T6}, the pixel value of the selected point is Tz, then the feature on one side = (Tz-T1)+(Tz-T2)+(Tz-T3), the feature on the other side = (Tz-T5)+(Tz-T6), and the interval value between the feature on one side and the feature on the other side is reconfirmed by confirmation, and the interval value = |feature on one side - feature on the other side|;

[0068] Confirm the feature midpoints of several equally divided lines in turn, connect the confirmed feature midpoints, confirm the feature line, and then use this feature line as the contour line of this contour area, and recalibrate the contour area (that is, the inner area of ​​the contour line is re-used as the corresponding contour area), completing the feature refinement process of this contour area;

[0069] The feature images confirmed in different feature image sets are refined in turn and displayed through specific display terminals for external medical staff to view.

[0070] Specifically, there are different contour areas in the feature image. In order to make the features most obvious, the corresponding contour areas are refined, and the feature midpoints are specifically confirmed based on the comprehensive performance of the pixels in the edge area. Then, based on the location of the feature midpoints, the edge contours of the corresponding contour areas are reconfirmed to make the edge features of the corresponding contour areas most obvious, thus completing the corresponding feature processing process.

[0071] Second embodiment

[0072] Combine Figure 2 , a real-time digestive endoscopy assistance system based on artificial intelligence, including:

[0073] The grayscale image conversion end performs grayscale processing on the microscopic images generated during the microscopic examination of the digestive endoscope, and confirms the grayscale image associated with the corresponding microscopic image;

[0074] On the feature classification side, the contour area in each frame of the microscopic image is confirmed based on the Sobel algorithm, and based on the overall contour features of the contour area, several frames of images are classified to confirm several groups of feature image sets;

[0075] The feature image calibration end verifies the area of ​​the contour region of the confirmed feature image set using different grayscale images, and selects the grayscale image with the largest contour region area as the feature image of the corresponding feature image set;

[0076] The contour feature refinement end refines the feature images confirmed in different feature image sets, confirms the contour line areas of different contour areas in the feature images, and selects feature midpoints from the contour line areas based on the pixel value change characteristics of the pixel points inside the contour line areas, and refines the features of the contour areas based on the locations of several feature midpoints.

[0077] Some of the data in the above formulas are dimensionless and numerically calculated. Meanwhile, the contents not described in detail in this specification belong to the prior art known to those skilled in the art.

[0078] The above embodiments are only used to illustrate the technical method of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. A real-time assisted method for digestive endoscopy based on artificial intelligence, characterized in that: The following steps are involved: Step 1: Process the microscopic images generated during the digestive endoscopy in real time, confirm the contour area in each frame of the microscopic image based on the Sobel algorithm, and classify several frames of images based on the overall contour features of the contour area to confirm several groups of feature image sets; Step 2: for the confirmed feature image set, different grayscale images are used to confirm the area of ​​the contour region, and the grayscale image with the largest contour region area is selected as the feature image of the corresponding feature image set; Step 3: Refine the features of the feature images confirmed in different feature image sets, confirm the contour line areas of different contour areas in the feature images, and select feature midpoints from the contour line areas based on the pixel value change characteristics of the pixel points inside the contour line areas, and refine the features of the contour areas based on the locations of several feature midpoints.

2. The method for real-time digestive endoscopy assistance based on artificial intelligence according to claim 1, characterized in that: In step 1, the specific method of confirming the contour area in each frame of the microscopic image is: Grayscale processing is performed on each frame of the microscopic image to confirm the grayscale image associated with the corresponding microscopic image; The Sobel algorithm is used to confirm the vertical gradient and vertical gradient associated with different pixels in the grayscale image, and the vertical gradient associated with the corresponding pixel is calibrated as C i , and the associated vertical gradient is calibrated as Z i , where i represents different pixels; use Confirm the comprehensive gradient ZH i , will satisfy: ZH i Pixels with Y1 > are calibrated as gradient pixels. Otherwise, no calibration is performed and Y1 is the preset value. Based on several groups of gradient pixel points confirmed in the grayscale image, continuous gradient pixel points are connected to confirm the contour line, and the middle area contained in the contour line is marked as the contour area.

3. The real-time assisted digestive endoscopy method based on artificial intelligence according to claim 2, characterized in that: In step 1, the specific method of classifying the multiple frames of images is as follows: Confirm the center point of the contour area associated with the grayscale image, and use the two-dimensional coordinate system to confirm the two-dimensional coordinates associated with different contour points in the contour line corresponding to the contour area. Then, perform mean processing on several groups of two-dimensional coordinates to confirm the mean coordinates. Based on the confirmed mean coordinates, perform center point calibration in the contour area. If there is only one set of contour areas in the grayscale image, the center point of the contour area is recorded as the feature point of the grayscale image; If there are two sets of contour areas in the grayscale image, the midpoint of the line connecting the center points of the two sets of contour areas is recorded as the feature point of the grayscale image; If there are more than two groups of contour areas in the grayscale image, connect the center points of the adjacent contour areas to determine a group of polygons, and record the center points of these polygons as the feature points of the grayscale image; The feature point positions of grayscale images of different frames that appear in sequence are confirmed. If the feature point position distance of adjacent frames of grayscale images is lower than the threshold, they are classified as the same type of grayscale images. If it is not lower than the threshold, the grayscale images of the same type are not calibrated, and multiple frames of different grayscale images belonging to the same type of grayscale images are divided into the same group of feature image sets.

4. The method for real-time digestive endoscopy assistance based on artificial intelligence according to claim 1, characterized in that: In step 2, the specific method of determining the feature image from the feature image set is: For different grayscale images in the feature image set, the contour areas confirmed in different grayscale images are confirmed, and the sum of the contour areas belonging to a single set of grayscale images is determined and calibrated as Mz k , where k represents different grayscale images; Then the total area parameter of the corresponding grayscale image is confirmed and calibrated as Zc k ; Use: Z k =Mz k ÷Zc k Confirm the area ratio feature Z of the corresponding contour area k , and then from several different area ratio features Z k , select Z k The grayscale image associated with max is used as the feature image of this feature image set.

5. The real-time assisted method for digestive endoscopy based on artificial intelligence according to claim 1, characterized in that: In step 3, based on the locations of several feature midpoints, the specific method of performing feature refinement of the contour area is as follows: Based on the contour line area associated with the contour area, connect the outermost gradient pixel points from the center point of the contour area to confirm the outer contour: based on the location of the center point of the area, determine the gradient pixel point farthest from the current position, and draw a circle with the radius of the center point of the area and this gradient pixel point. Then, sequentially determine the gradient pixel points closest to the circumference in the contour line area and record them as outer contour points. Connect several outer contour points to confirm the outer contour; Then connect the innermost gradient pixel points from the center point of the contour area to confirm the internal contour: determine the gradient pixel point closest to the current position based on the location of the center point of the area, and draw a circle with the radius of the center point of the area and the gradient pixel point. Then, sequentially determine the gradient pixel points closest to the circumference of the circle in the contour connection area and record them as internal contour points. Connect several internal contour points to confirm the internal contour; Based on the confirmed outer contour and inner contour, contour points are sequentially selected and connected in the outer contour and the inner contour to determine a number of bisectors, wherein the bisectors divide the contour connection area into n areas of equal area, where n is a preset value; Identify the pixels associated with each set of bisectors and mark them as the pending pixels of this bisector. Then select the equilibrium point among several pending pixels and record it as the feature midpoint. Confirm the feature midpoints of several equally divided lines in turn, connect the confirmed feature midpoints, confirm the feature connection line, and then use this feature connection line as the contour connection line of this contour area, and recalibrate the contour area to complete the feature refinement process of this contour area; The feature images confirmed in different feature image sets are refined in turn and displayed through specific display terminals.

6. The real-time assisted method for digestive endoscopy based on artificial intelligence according to claim 5, characterized in that: The specific method of selecting a balance point from among several undetermined pixel points and recording it as the feature midpoint is as follows: Sort the pixel values ​​of several pending pixels from one end point of the bisector to the other end point, and confirm whether the sorted pixel values ​​show a gradually increasing or decreasing state: If yes, then the maximum value is selected from the sorted multiple pixel values, and the undetermined pixel point corresponding to the maximum value is marked as the feature midpoint; If not, then select a point from the middle pending pixel points included in the bisector, record this point and several pending pixel points between the endpoint on one side as one side pixel point set, record this point and several pending pixel points between the endpoint on the other side as the other side pixel point set, calculate the difference between the pixel value of this point and the pixel value of the pixel point set on one side in turn, and determine the sum of the differences and record it as one side feature, then calculate the difference between the pixel value of the point and the pixel value of the pixel point set on the other side in turn, and determine the sum of the differences and record it as the other side feature, confirm the interval value between the feature on one side and the feature on the other side, and use this interval value as the process feature of the selection process of this point, then select other middle pending pixel points in turn, and determine the process feature of the corresponding selected process, lock the pending pixel point associated with the minimum process feature from several process features and record it as the feature midpoint.

7. The real-time assisted method for digestive endoscopy based on artificial intelligence according to claim 6, characterized in that: The pixel point set on one side includes an endpoint on one side, and the pixel point set on the other side includes an endpoint on the other side.

8. A real-time digestive endoscopy assistance system based on artificial intelligence, which operates based on a real-time digestive endoscopy assistance method based on artificial intelligence according to any one of claims 1 to 7, characterized in that: include: The grayscale image conversion end performs grayscale processing on the microscopic images generated during the microscopic examination of the digestive endoscope, and confirms the grayscale image associated with the corresponding microscopic image; On the feature classification side, the contour area in each frame of the microscopic image is confirmed based on the Sobel algorithm, and based on the overall contour features of the contour area, several frames of images are classified to confirm several groups of feature image sets; The feature image calibration end verifies the area of ​​the contour region of the confirmed feature image set using different grayscale images, and selects the grayscale image with the largest contour region area as the feature image of the corresponding feature image set; The contour feature refinement end refines the feature images confirmed in different feature image sets, confirms the contour line areas of different contour areas in the feature images, and selects feature midpoints from the contour line areas based on the pixel value change characteristics of the pixel points inside the contour line areas, and refines the features of the contour areas based on the locations of several feature midpoints.

Images