Image value analysis method and system for massive images in alimentary canal

Abstract

The invention relates to the technical field of image recognition, and discloses an image value analysis method and system for a large number of images in an alimentary canal, and the method comprises the steps: obtaining an alimentary canal image sequence, calculating an optical flow field between adjacent frames, and classifying the alimentary canal image sequence based on the rotation and divergence of the optical flow field, obtaining a motion type label of the alimentary canal image sequence; performing redundancy processing on the alimentary canal image sequence by using the optical flow field to obtain a screened alimentary canal image sequence; performing feature extraction on the screened alimentary canal image sequence, determining an anatomical structure turning point of the screened alimentary canal image sequence and labeling an anatomical position label; carrying out image value analysis on the screened alimentary canal image sequence according to a motion type label, and identifying a lesion area; and carrying out value scoring on the screened alimentary canal image sequence based on the lesion area and the anatomical position label to obtain a value analysis result. And the analysis efficiency and accuracy of the digestive tract image are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image recognition, in particular to an image value analysis method and system for a large number of images in the digestive tract. BACKGROUND

[0002] With the popularization of medical imaging technologies such as capsule endoscopy and high-definition electronic endoscopy, the amount of image data generated by digestive tract examination has exploded, forming a massive video data stream. Traditional manual film reading by physicians faces challenges such as low efficiency, fatigue, and high missed diagnosis rate. Therefore, developing computer-aided diagnosis technology to automatically analyze, eliminate redundancy, and evaluate the value of massive digestive tract images to assist physicians in quickly locating high-value lesions or key anatomical structure images has become a research hotspot in the field of intelligent medicine. Existing technologies mainly focus on redundancy elimination based on inter-frame similarity, lesion preliminary detection based on texture or color features, and image quality evaluation and rough anatomical position classification based on deep learning.

[0003] However, in the face of the complex environment inside the digestive tract and the multi-dimensional analysis needs of massive data, the intelligence and refinement of existing methods are still insufficient. First, the inherent physiological peristalsis and non-physiological lens movement coexist in the digestive tract, resulting in complex and variable motion states of image sequences. Existing methods are difficult to effectively identify and classify such complex motion, resulting in insufficient analysis accuracy and stability in sequence processing. Second, in the aspect of redundancy elimination, traditional methods often use fixed similarity thresholds, lack adaptability to different motion states, and are prone to loss of key lesion frames or excessive retention of invalid static frames. Third, existing lesion detection strategies lack specificity, often using a single model to process all images, ignoring the differences in lesion feature performance under static and peristaltic states, and limiting the robustness of detection. Finally, the image value evaluation dimension is single, mostly limited to the presence or absence of lesions, and fails to effectively integrate anatomical position and other contextual information, making it difficult to accurately rank the comprehensive clinical value of images. SUMMARY

[0004] In view of the above problems, the present application is proposed.

[0005] To solve the above technical problems, the present application provides the following technical solutions: an image value analysis method for a large number of images in the digestive tract, comprising:

[0006] acquiring a digestive tract image sequence and calculating the optical flow field between adjacent frames, classifying the digestive tract image sequence based on the vorticity and divergence of the optical flow field, and obtaining the motion type label of the digestive tract image sequence; the motion type label includes stable images, peristaltic images, and non-peristaltic images;

[0007] The gastrointestinal image sequence is redundantly processed by using the optical flow field to obtain a screened gastrointestinal image sequence;

[0008] Anatomical structure turning points of the screened gastrointestinal image sequence are determined and anatomical position labels are labeled.

[0009] An image value analysis is performed on the screened gastrointestinal image sequence according to the motion type label to identify a lesion region; the image value analysis comprises analyzing the stable type image to identify a candidate lesion region, performing motion characteristic analysis on the candidate lesion region in combination with the peristaltic type image, and determining a lesion attribute of the candidate lesion region according to a result of the motion characteristic analysis.

[0010] A value score is given to the screened gastrointestinal image sequence based on the lesion region and the anatomical position label to obtain a value analysis result.

[0011] As a preferred scheme of the image value analysis method for a large number of gastrointestinal internal images, the motion type label of the gastrointestinal image sequence is obtained by classifying the gastrointestinal image sequence based on the curl and divergence of the optical flow field, which comprises calculating based on the horizontal component and the vertical component of the optical flow field to generate a divergence field and a curl field.

[0012] The divergence field and the curl field are globally spatially aggregated to calculate a global motion feature representing the overall motion amplitude of the image sequence, and the gastrointestinal image sequence is preliminarily divided by using the global motion feature to obtain stable type images and motion type images.

[0013] Pixel points of the divergence field corresponding to the motion type image are identified, and the identified pixel points are clustered to obtain a divergence significant region.

[0014] Optical flow vectors in the divergence significant region are extracted, and a direction distribution of the optical flow vectors is analyzed to obtain a direction consistency feature of the optical flow vectors, the motion type image is secondarily divided according to the direction consistency feature to obtain peristaltic type images and non-peristaltic type images, and the motion type label is labeled.

[0015] As a preferred scheme of the image value analysis method for a large number of gastrointestinal internal images, the screened gastrointestinal image sequence is obtained by redundantly processing the gastrointestinal image sequence by using the optical flow field, which comprises inversely mapping a current frame by using optical flow vectors in the optical flow field to obtain an aligned frame after motion compensation, and calculating a pixel difference degree between the aligned frame and a previous frame.

[0016] The corresponding redundancy threshold is obtained based on the motion type label of the current frame; among them, the first redundancy threshold is used for stable images, the second redundancy threshold is used for creeping images, and no redundancy threshold is set for non-creeping images, and all frames are directly removed;

[0017] The pixel difference is compared with the corresponding redundancy threshold. When the pixel difference is less than or equal to the redundancy threshold, the current frame is determined to be a redundant frame and is removed. When the pixel difference is greater than the redundancy threshold, the current frame is retained, and the filtered digestive tract image sequence is obtained.

[0018] As a preferred embodiment of the image value analysis method for a large number of images of the digestive tract as described in this invention, the method includes: extracting features from the selected digestive tract image sequence, determining the anatomical inflection points of the selected digestive tract image sequence and labeling the anatomical locations, and extracting color features, texture features and edge features from each image in the selected digestive tract image sequence and combining them to form a feature vector.

[0019] The feature vector is reduced in dimensionality and mapped to the feature space to form a sequence of feature points. The Euclidean distance between adjacent feature points is then calculated.

[0020] The locations of abrupt changes in the Euclidean distance are detected to obtain abrupt change points, and the images corresponding to the abrupt change points are marked as anatomical structural turning points.

[0021] Based on the location of the inflection points of the anatomical structures, the image segments between adjacent inflection points are labeled with the corresponding anatomical location tags.

[0022] As a preferred embodiment of the image value analysis method for a large number of images of the digestive tract as described in this invention, the analysis of the stable images and the identification of candidate lesion areas include: using a sliding window to traverse the filtered digestive tract image sequence corresponding to the stable images, calculating the information entropy of the pixel grayscale distribution within the sliding window and using it as the texture entropy value.

[0023] For the sliding window, the degree of deviation of the texture entropy value of the sliding window relative to the texture entropy of the neighborhood is calculated, and the position where the deviation exceeds the judgment criterion is marked as the entropy mutation point;

[0024] The entropy mutation markers at each image location are statistically analyzed, and the locations marked as entropy mutations are identified as candidate lesion regions.

[0025] As a preferred embodiment of the image value analysis method for a large number of images of the digestive tract as described in this invention, the following steps are taken: combining the peristaltic images to perform motion characteristic analysis on the candidate lesion regions, and determining the lesion attributes of the candidate lesion regions based on the results of the motion characteristic analysis, including tracing the candidate lesion regions in the filtered digestive tract image sequence corresponding to the peristaltic images;

[0026] The optical flow field is used to extract the local optical flow vector of the candidate lesion region and the background optical flow vector of the neighboring background region;

[0027] The motion difference between the local optical flow vector and the background optical flow vector is calculated, and the lesion attributes of the candidate lesion region are determined based on the motion difference metric; the lesion attributes include high-rigidity lesions, low-rigidity tissues, and motion artifacts.

[0028] As a preferred embodiment of the image value analysis method for a large number of images of the digestive tract as described in this invention, the following steps are taken: The value of the selected digestive tract image sequence is scored based on the lesion region and the anatomical location label to obtain the value analysis result. This includes: for each image in the selected digestive tract image sequence, an initial value score is assigned based on the lesion region; wherein images containing high-rigidity lesions are assigned a high initial score, images containing low-rigidity tissues are assigned a medium initial score, and images not containing lesion regions or only containing motion artifacts are assigned a low initial score.

[0029] Based on the anatomical location labels, the temporal distance between each image and the corresponding anatomical structure inflection point is calculated, and the initial value score is adjusted according to the temporal distance to obtain the comprehensive value score of each image.

[0030] The selected digestive tract image sequences are arranged according to the comprehensive value score, and the sorted digestive tract image sequences are used as the value analysis results.

[0031] An image value analysis system for a large number of images of the digestive tract interior, wherein:

[0032] The image classification module acquires a sequence of digestive tract images and calculates the optical flow field between adjacent frames. Based on the curl and divergence of the optical flow field, it classifies the digestive tract image sequence to obtain motion type labels for the digestive tract image sequence. The motion type labels include stable images, peristaltic images, and non-peristaltic images.

[0033] The redundancy processing module uses the optical flow field to perform redundancy processing on the digestive tract image sequence to obtain a filtered digestive tract image sequence.

[0034] The anatomical location module extracts features from the filtered digestive tract image sequence, determines the anatomical inflection points of the filtered digestive tract image sequence, and labels the anatomical locations.

[0035] The lesion identification module performs image value analysis on the screened digestive tract image sequence based on the motion type label to identify lesion areas; the image value analysis includes analyzing the stable images to identify candidate lesion areas, combining the peristaltic images to perform motion characteristic analysis on the candidate lesion areas, and determining the lesion attributes of the candidate lesion areas based on the results of the motion characteristic analysis.

[0036] The image value module scores the value of the selected digestive tract image sequences based on the lesion area and the anatomical location label to obtain the value analysis results.

[0037] A computer device includes: a memory and a processor; the memory stores a computer program, wherein: when the processor executes the computer program, it implements the steps of the method described in any one of the present invention.

[0038] A computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the method described in any one of the present invention.

[0039] The beneficial effects of this invention are as follows: The image value analysis method for a large number of images of the digestive tract provided by this invention effectively classifies the motion state of digestive tract image sequences by calculating the curl and divergence of the optical flow field. In the redundancy removal stage, by introducing a threshold matching the motion type label, the image sequences are filtered to ensure that key lesion frames are not removed due to changes in motion state while compressing data, thus achieving a balance between data simplification and information fidelity. In the lesion identification stage, by utilizing the rigidity characteristics of lesion tissue and the differences in motion during peristalsis, the method accurately distinguishes between high-rigidity lesions, low-rigidity tissues, and motion artifacts, improving the accuracy of lesion identification and reducing the misdiagnosis and missed diagnosis rates. In the value assessment stage, by annotating anatomical structural inflection points through manifold analysis and integrating lesion attributes and anatomical location labels for comprehensive scoring, the output value analysis results are more accurate, improving the analysis efficiency and accuracy of digestive tract images. Attached Figure Description

[0040] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0041] Figure 1 The first embodiment of the present invention provides an overall flowchart of an image value analysis method for a large number of images of the digestive tract. Detailed Implementation

[0042] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0043] Example 1, referring to Figure 1 As an embodiment of the present invention, an image value analysis method for a large number of images of the digestive tract is provided, comprising:

[0044] S1: Acquire a sequence of digestive tract images and calculate the optical flow field between adjacent frames. Classify the digestive tract image sequence based on the curl and divergence of the optical flow field to obtain motion type labels for the digestive tract image sequence. The motion type labels include stable images, peristaltic images, and non-peristaltic images.

[0045] In this embodiment, a digestive tract image sequence is first acquired, which can be acquired by medical imaging devices such as capsule endoscopy and electronic endoscopy. The digestive tract image sequence contains thousands to tens of thousands of consecutive images. These images record the movement process of the medical imaging device inside the digestive tract. Due to the natural peristalsis of the digestive tract, coupled with the movement and rotation of the medical imaging device itself, there are complex motion relationships between the images.

[0046] Furthermore, for the acquired digestive tract image sequence, the optical flow method is used to calculate the optical flow field between adjacent frames. The optical flow field is used to describe the motion vector distribution of each pixel in the image between adjacent frames. In this embodiment, the dense optical flow algorithm is used to calculate the optical flow field for the t-th frame and the (t+1)-th frame in the digestive tract image sequence. The optical flow field contains the horizontal component of each pixel. and vertical components ,in Represents the spatial coordinates of a pixel. Represents the horizontal coordinate. Represents the coordinates in the vertical direction.

[0047] Specifically, based on the horizontal component of the optical flow field and vertical components The divergence field and curl field are calculated separately. The divergence field reflects the divergence or convergence characteristics of the optical flow vector. A positive divergence value indicates expansion in the corresponding region, while a negative divergence value indicates contraction. The curl field reflects the rotational characteristics of the optical flow vector. The larger the absolute value of the curl, the more pronounced the rotational motion. In this embodiment, based on the differential operator theory of vector fields, the divergence field is obtained by taking the partial derivatives of the horizontal component of the optical flow vector with respect to the horizontal direction and the partial derivatives of the vertical component with respect to the vertical direction, and then summing them. The curl field is obtained by calculating the difference between the partial derivatives of the vertical component with respect to the horizontal direction and the partial derivatives of the horizontal component with respect to the vertical direction. Through the combined analysis of divergence and curl, the characteristics of tissue motion in the digestive tract image can be comprehensively characterized, providing a quantitative basis for subsequent motion type classification.

[0048] Furthermore, based on the divergence field and curl field, the digestive tract image sequence is classified, including the following steps: global spatial aggregation of the divergence field and curl field, and calculation of global motion features characterizing the overall motion amplitude of the image sequence. The global motion features include the global divergence mean and the global curl mean, and the specific formulas are as follows:

[0049] ;

[0050] ;

[0051] in, This represents the mean of the global divergence. This represents the total number of pixels in the image; This represents the divergence value of the i-th pixel. This represents the global curl mean; This represents the curl value of the i-th pixel.

[0052] The global divergence mean and global curl mean Each with a preset divergence threshold Curl threshold When comparing, the global divergence mean Less than or equal to the divergence threshold And global curl mean Less than or equal to the curl threshold When the overall motion amplitude in the image is small, the corresponding frame is classified as a stable image; when the global divergence mean is small... Greater than the divergence threshold or global curl mean greater than the curl threshold When the image is determined to have obvious motion, the corresponding frame is initially classified as a moving image.

[0053] Among them, for the divergence threshold Curl threshold By randomly selecting several frames from the acquired digestive tract image sequence as training samples, and manually labeling the stable images that are obviously stationary or slowly moving, the mean global divergence and mean global curl of these stable images are calculated and their distribution is statistically analyzed. The mean plus one standard deviation is taken as the stability threshold. .

[0054] It should be noted that using both divergence and curl as indicators for stability determination can simultaneously capture expansion or contraction motion (divergence) and rotational motion (curl) in the image. Compared with using only a single motion indicator, this improves the accuracy of stable image recognition and avoids misjudging local minute motions as overall stability.

[0055] Furthermore, for moving images, traditional methods often discard all moving images as low-quality images. However, the physiological peristaltic movements of the digestive tract are fundamentally different from non-physiological movements such as random shaking and rapid sliding of equipment. Peristaltic movements are orderly and directional tissue contractions or relaxations, while non-peristaltic movements are characterized by chaotic and disordered movement patterns. Therefore, this invention performs a secondary judgment on moving images.

[0056] Specifically, for digestive tract image sequences initially classified as motion images, further determination of whether the corresponding digestive tract image sequence belongs to peristaltic or non-peristaltic images includes the following steps: identifying significant divergence regions in the divergence field, and identifying regions with absolute divergence values ​​greater than a preset divergence threshold. The pixels are clustered to obtain the divergence salient region Ω, which corresponds to the most turbulent part of the image and is the main area of ​​peristaltic motion. By calculating the statistical distribution of the divergence field of the current image, the mean and standard deviation of the absolute divergence value are obtained. The divergence threshold is obtained by combining the mean and standard deviation of the absolute divergence value. .

[0057] For each significant divergence region Ω, all optical flow vectors within Ω are extracted, and their directional distribution is analyzed to obtain the directional consistency characteristics. Specifically, the orientation angles of each optical flow vector are calculated, and a histogram of orientation angle distribution is plotted. Based on the histogram, the dominant direction of the optical flow vectors (i.e., the direction interval with the largest number of optical flow vectors) is identified. The proportion of the dominant direction optical flow vectors to the total optical flow vectors is calculated to obtain the directional consistency characteristics. ; directional consistency feature Consistency threshold with preset Compare them. When the directional consistency feature... Greater than the consistency threshold When the optical flow vectors within the significant divergence region exhibit high directional consistency, they show a consistent flow along the long axis of the digestive tract, consistent with the orderly motion characteristics of peristalsis, thus classifying the corresponding frame as a peristaltic image; when the directional consistency characteristic... Less than or equal to the consistency threshold If the optical flow vector direction is dispersed, it does not conform to the orderliness of peristaltic motion. The image of the corresponding frame is determined to be a non-peristaltic image, and the motion type label is completed.

[0058] It should be noted that although peristaltic images exhibit motion, this motion is an ordered physiological peristalsis. In subsequent lesion analysis, the deformation information generated by peristalsis can be used to detect abnormal tissue stiffness, thereby identifying early lesions. In contrast, non-peristaltic images, due to their chaotic and disordered motion, have poor image quality and do not contain valuable diagnostic information. The typical characteristic of peristaltic images is the ordered contraction or relaxation of the digestive tract wall, which manifests as a concentrated distribution of optical flow vectors in a specific direction. By extracting regions with significant divergence and analyzing the consistency of internal optical flow directions, peristaltic motion can be effectively distinguished from non-peristaltic motions such as random jittering and rapid sliding. Compared to traditional methods for determining motion amplitude, this invention improves the recognition accuracy of peristaltic images and reduces the misjudgment rate of non-peristaltic images by introducing directional consistency features.

[0059] In an optional embodiment, principal component analysis can also be used to determine the principal direction of the optical flow vector. This involves performing principal component analysis on the vector set consisting of all optical flow vectors within the divergence region. The direction of the first principal component is the principal direction. Specifically, the ratio of the projected length of all optical flow vectors in the direction of the first principal component to their own magnitude is calculated, and the average of all ratios is used as a directional consistency feature. This allows for more accurate capture of the dominant direction of the optical flow field, accurately distinguishing the images of the three motion types and laying the foundation for subsequent differential processing.

[0060] S2: The digestive tract image sequence is subjected to redundancy processing using the optical flow field to obtain a filtered digestive tract image sequence.

[0061] In this embodiment, the digestive tract image sequence is redundantly processed based on the optical flow field and motion type labels calculated in S1. Since there are a large number of similar or repeated frames in the digestive tract image sequence, these redundant frames are mainly caused by two reasons: first, multiple similar images are continuously taken when the capsule stays or moves slowly in a certain area; second, the peristalsis of the digestive tract causes the same tissue area to appear repeatedly in different frames but with slightly different perspectives. Directly processing all images would lead to a waste of computing resources. Therefore, it is necessary to remove redundant frames and retain key frames with diagnostic value.

[0062] Specifically, for the current frame, the reverse mapping operation using the optical flow vector in the optical flow field includes mapping the pixels in the current frame back to the corresponding positions in the previous frame according to the direction and amplitude of the optical flow vector, thereby obtaining the motion-compensated aligned frame. Since the mapped position may not be an integer coordinate, an interpolation method is needed to obtain the pixel value at the corresponding position. In this embodiment, a bilinear interpolation method is used to calculate the pixel value at the non-integer coordinate position by weighted averaging of four adjacent integer coordinate pixels. The aligned frame eliminates the positional offset between adjacent frames caused by camera movement or tissue crease, making adjacent frames spatially aligned, which facilitates subsequent comparison of pixel differences.

[0063] Furthermore, the pixel difference between the aligned frame and the previous frame is calculated. The pixel difference is obtained by comparing the grayscale values ​​of the aligned frame and the previous frame pixel by pixel and summing the differences of all pixels. The pixel difference measures the amount of information change of the current frame relative to the previous frame. The smaller the pixel difference, the more similar the current frame is to the previous frame and the higher the redundancy.

[0064] Furthermore, different redundancy judgment criteria are set for images with different motion types; specifically, for stable images, since the image content changes slowly, a first redundancy threshold is set. The first redundancy threshold is relatively low and can effectively remove highly similar frames in stable scenes; for creeping images, due to tissue movement, the image content changes relatively quickly, so a second redundancy threshold is set. The second redundancy threshold is relatively high to avoid normal changes caused by peristaltic movement being misjudged as redundancy. The first redundancy threshold... Less than the second redundancy threshold For non-creeping images, since these images usually correspond to abnormal motion states such as rapid sliding and violent shaking, the image quality is poor and lacks diagnostic value. Therefore, no redundant threshold is set, and all non-creeping image frames are directly removed.

[0065] When the pixel difference is less than or equal to the redundancy threshold, the current frame is determined to be a redundant frame and is removed; when the pixel difference is greater than the redundancy threshold, the current frame is determined to contain enough new information and is retained; all retained image frames are summarized to obtain the filtered digestive tract image sequence. The filtered digestive tract image sequence removes redundant frames and non-peristaltic image frames, retains key frames with diagnostic value, and avoids a large amount of redundancy caused by peristalsis.

[0066] It should be noted that the design of using categorical redundancy thresholds can adopt differentiated screening strategies for images with different motion characteristics, avoiding over-screening or under-screening caused by uniform thresholds, and retaining key diagnostic information to the maximum extent while reducing the amount of data.

[0067] S3: Extract features from the filtered digestive tract image sequence, determine the anatomical inflection points of the filtered digestive tract image sequence, and label the anatomical locations.

[0068] In this embodiment, since the digestive tract includes multiple anatomical regions such as the esophagus, stomach, duodenum, small intestine, and colon, and the image features of different regions are different, by extracting features and annotating anatomical locations from the selected digestive tract image sequence, the turning points of anatomical structures can be identified, thereby automatically dividing anatomical regions and annotating location information, providing anatomical basis for subsequent lesion localization.

[0069] Specifically, multi-dimensional features are extracted from each image in the selected digestive tract image sequence, including color features, texture features, and edge features. Color features are obtained by calculating the color histogram or color moments of the image in different color spaces, reflecting the color distribution characteristics of the image. Texture features are obtained by calculating texture descriptors such as the gray-level co-occurrence matrix and local binary mode, reflecting the texture roughness and directionality of the image. Edge features are obtained by extracting the edge distribution of the image and statistically analyzing the edge density and orientation distribution through edge detection operators, reflecting the structural information of the image. The above color features, texture features, and edge features are combined to form a high-dimensional feature vector, with one feature vector corresponding to each image.

[0070] Because eigenvectors have high dimensionality, direct analysis in high-dimensional space is computationally intensive and susceptible to the curse of dimensionality. Therefore, dimensionality reduction is performed on eigenvectors to map them to a low-dimensional feature space. Specifically, principal component analysis can be used to reduce dimensionality by calculating the eigenvalues ​​and eigenvectors of the eigenvector covariance matrix and extracting the principal components with the largest variance, thus achieving dimensionality reduction while retaining key information. After dimensionality reduction, each image corresponds to a feature point in the feature space. Arranged chronologically according to the images in the sequence, a feature point sequence is formed. This feature point sequence constitutes a trajectory in the feature space, reflecting the evolution of digestive tract image features.

[0071] Furthermore, the Euclidean distance between adjacent feature points in the feature point sequence is calculated to quantify the similarity between adjacent images in the feature space. The larger the Euclidean distance, the greater the feature difference between adjacent images. During the gastrointestinal examination, when the camera moves from one anatomical area to another, the image features will change abruptly, which is manifested as an increase in the distance between adjacent feature points in the feature point sequence.

[0072] Secondly, mutation locations in the Euclidean distance sequence are detected. The detection of mutation locations can be achieved by using the mean within a sliding window. When the Euclidean distance at a certain location is greater than the average distance of the neighborhood, the corresponding location is determined to be a mutation point. The image corresponding to the mutation point is marked as an anatomical structural inflection point, which marks the boundary of the digestive tract anatomical region.

[0073] Based on the location of the anatomical inflection points, the image segments between adjacent inflection points are divided into an anatomical region and labeled with corresponding anatomical location tags. The anatomical location tags can be inferred based on the number and order of the inflection points. For example, the first segment is usually the esophagus, the second segment is the stomach, the third segment is the duodenum, and so on. In an optional embodiment, the anatomical location tags can also be verified and corrected by combining the clustering results of image features and medical prior knowledge to further improve the labeling accuracy.

[0074] It should be noted that by combining manifold learning and distance mutation detection to identify anatomical inflection points, this invention fully utilizes the manifold distribution characteristics of digestive tract images in feature space. Compared with traditional segmentation methods based on a single feature threshold, this invention can more accurately capture the boundaries between anatomical regions, adapt to the differences in image features between different individuals and different examination equipment, and improve the robustness and accuracy of anatomical location annotation.

[0075] S4: Perform image value analysis on the screened digestive tract image sequence according to the motion type label to identify lesion areas; the image value analysis includes analyzing the stable images to identify candidate lesion areas, combining the peristaltic images to perform motion characteristic analysis on the candidate lesion areas, and determining the lesion attributes of the candidate lesion areas based on the results of the motion characteristic analysis.

[0076] In this embodiment, image value analysis is performed on the screened digestive tract image sequences based on motion type labels. Identifying lesion areas means leveraging the complementary advantages of stable and peristaltic images: stable images are of high quality and clear, suitable for fine texture analysis to initially identify candidate lesion areas; peristaltic images contain tissue motion information, and by analyzing the motion characteristics of the lesion area during peristalsis, the rigidity of the lesion can be further determined, thereby distinguishing real lesions from motion artifacts.

[0077] Furthermore, the specific steps for image value analysis of the screened digestive tract image sequences based on the motion type labels include:

[0078] The images labeled as stable are analyzed to identify candidate lesion regions. In this embodiment, the multi-scale texture entropy detection method is used to identify candidate lesion regions, which includes: using multiple sliding windows of different scales to traverse the filtered digestive tract image sequence corresponding to the stable images, and calculating the information entropy of the pixel gray-level distribution within each sliding window as the texture entropy value. The information entropy reflects the degree of disorder in the gray-level distribution. The texture of normal digestive tract tissue is relatively uniform, and the information entropy is within the normal range. However, due to abnormal tissue structure, the texture of lesion regions is disordered, and the information entropy often deviates from the normal range.

[0079] For each sliding window position, the deviation of the corresponding texture entropy value from the neighborhood texture entropy is calculated. The neighborhood texture entropy is obtained by statistically analyzing the texture entropy values ​​of several neighboring windows around the current window and calculating the average value. The deviation is obtained by calculating the absolute value of the difference between the texture entropy value of the current window and the average texture entropy of the neighborhood, and then normalizing it after calculating the standard deviation of the neighborhood texture entropy. The greater the deviation, the more obvious the difference between the texture features of the current window and the surrounding normal area, and the more likely it is to be a lesion area.

[0080] Furthermore, the degree of deviation is compared with a preset judgment standard. When the degree of deviation exceeds the judgment standard, the window position is marked as an entropy mutation point. In this embodiment, the judgment standard can be determined based on the statistical distribution of historical data, and is usually set to several times the standard deviation. By detecting entropy mutation points at multiple scales, texture anomalies can be identified at different spatial resolutions.

[0081] For each location in the image, if the corresponding location is marked as an entropy mutation point at multiple scales, it indicates that the texture anomaly at the corresponding location has multi-scale consistency and is more likely to be a real lesion. Therefore, the location marked as an entropy mutation at multiple scales is identified as a candidate lesion region, and morphological processing is performed on the candidate lesion region, including region connectivity analysis and boundary extraction, to obtain the contour and location information of the candidate lesion region.

[0082] It should be noted that the multi-scale texture entropy detection method can effectively identify abnormal texture regions in digestive tract images. Multi-scale analysis improves the robustness of detection and avoids false detections or missed detections at a single scale. By calculating the degree of deviation of texture entropy relative to the neighborhood, it can adapt to the normal texture features of different individuals and different regions, thereby improving the accuracy of candidate lesion region identification.

[0083] Furthermore, since candidate lesion areas appear as texture abnormalities in stable images, but the cause of texture abnormalities may be real lesions, motion artifacts (such as bubbles), or local variations in normal tissue, it is necessary to utilize motion information in peristaltic images to further determine the nature of candidate lesion areas. Due to their rigid texture and reduced elasticity, lesion tissue undergoes smaller deformation during peristaltic movement, exhibiting high rigidity characteristics. In contrast, normal tissue has good elasticity and undergoes larger deformation during peristalsis, exhibiting low rigidity characteristics. Motion artifacts may exhibit abnormal motion patterns inconsistent with the surrounding tissues.

[0084] Specifically, the process of analyzing the motion characteristics of candidate lesion regions using peristaltic images and determining the lesion attributes of candidate lesion regions based on the results of the motion characteristic analysis includes the following steps: Tracking candidate lesion regions in the filtered digestive tract image sequence corresponding to the peristaltic images. The purpose of tracking is to locate the position of candidate lesion regions in consecutive peristaltic image frames to analyze their motion characteristics. In this embodiment, firstly, based on the position information of the candidate lesion region in the stable image, the corresponding region is found in the peristaltic image that is closest in time. Since the stable image and the peristaltic image may have temporal intervals, it is necessary to combine anatomical location labels and image feature similarity for matching. For the matched peristaltic image, the optical flow field calculated in S1 is used to track the boundary of the candidate lesion region in consecutive peristaltic image frames. Guided by the optical flow vector, the contour of the candidate lesion region is mapped from the current frame to the next frame, achieving cross-frame tracking.

[0085] Furthermore, the extraction of local optical flow vectors from the candidate lesion region and background optical flow vectors from the neighboring background region using optical flow fields includes: local optical flow vectors refer to the set of optical flow vectors corresponding to all pixels within the candidate lesion region, reflecting the motion characteristics of the lesion region itself. Background optical flow vectors refer to the set of optical flow vectors from the normal tissue region surrounding the candidate lesion region, reflecting the motion characteristics of the background tissue. In this embodiment, the neighboring background region is defined as a ring-shaped region extending outward from the candidate lesion region at a certain distance. The ring-shaped region excludes the candidate lesion region itself and only includes the surrounding normal tissue. By extracting the optical flow vectors within the neighboring background region, the set of background optical flow vectors is obtained.

[0086] Furthermore, the motion difference between the local optical flow vector and the background optical flow vector is calculated as follows: First, the average amplitude of the local optical flow vector and the average amplitude of the background optical flow vector are calculated separately; second, the absolute value of the difference between the local average amplitude and the background average amplitude is calculated as the main component of the motion difference; in addition, the consistency difference of the optical flow vector direction can also be considered, and the difference in the principal direction angle between the local optical flow vector and the background optical flow vector is calculated as an auxiliary component of the motion difference; the motion difference is obtained by combining the difference in motion amplitude and the difference in direction.

[0087] Finally, the lesion attributes of the candidate lesion area are determined based on the motion difference. The lesion attributes include high rigidity lesions, low rigidity tissues, and motion artifacts. The determination logic is as follows: when the motion difference is large and the average amplitude of the local optical flow vector is less than the average amplitude of the background optical flow vector, it indicates that the motion amplitude of the candidate lesion area during the peristalsis process is significantly smaller than that of the surrounding normal tissue, indicating that the corresponding area has high rigidity characteristics and is not easily deformed by peristalsis.

[0088] When the motion difference is small and the average amplitude of the local optical flow vector is equal to the average amplitude of the background optical flow vector, it indicates that the motion amplitude of the candidate lesion area and the surrounding tissue is similar during peristalsis, indicating that the corresponding area has normal elasticity and can deform synchronously with peristalsis. The candidate lesion area is determined to be low-rigidity tissue and is therefore a secondary focus.

[0089] When the motion difference is large, but the local optical flow vector shows a motion pattern completely different from the background optical flow vector (such as opposite direction, abnormally large amplitude, etc.), it indicates that the motion characteristics of the candidate lesion area are significantly inconsistent with the surrounding tissue and do not conform to the law of tissue rigidity difference. Such features usually correspond to motion artifacts, such as the drift of bubbles during peristalsis, foreign objects in front of the lens, etc. The candidate lesion area is judged as a motion artifact, which has no diagnostic value and is directly eliminated.

[0090] In an optional embodiment, the temporal stability characteristics of candidate lesion regions in continuous creeping image frames can be combined to further verify the lesion attribute determination results. For example, for regions determined to be high-rigidity lesions, their morphology and position in continuous frames should show high stability; for regions determined to be motion artifacts, their morphology may change rapidly or their position may drift significantly in continuous frames. Through temporal stability analysis, the accuracy of lesion attribute determination can be improved.

[0091] It should be noted that by combining texture analysis of stable images and motion analysis of peristaltic images, this invention fully utilizes the complementary advantages of the two types of images. Stable images provide high-quality spatial texture information for preliminary screening of candidate lesion areas; peristaltic images provide dynamic motion information for determining the rigidity of candidate lesion areas, thereby distinguishing between real lesions and artifacts. Peristalsis is transformed from a traditional interference factor into a diagnostic tool. By utilizing the rigidity characteristics of lesion tissue during peristalsis, accurate identification and attribute determination of lesions are achieved. Compared with texture analysis relying solely on static images or temporal analysis relying solely on moving images, this invention improves the accuracy of lesion identification and reduces the misdiagnosis and missed diagnosis rates.

[0092] S5: The value score of the screened digestive tract image sequence is obtained based on the lesion area and the anatomical location label to obtain the value analysis results.

[0093] Further, the selected digestive tract image sequences are value-scored based on the lesion region and the anatomical location label to obtain value analysis results, including the following steps:

[0094] For each image in the selected digestive tract image sequence, an initial value score is assigned based on the lesion area identified by S4. Images containing high-rigidity lesions are assigned a high initial score, images containing low-rigidity tissues are assigned a medium initial score, and images containing no lesion area or only motion artifacts are assigned a low initial score.

[0095] It should be noted that the initial value score reflects the diagnostic value of the image in terms of lesion information, and the scoring rules are set based on the clinical significance of the lesion attributes. High-rigidity lesions usually correspond to severe pathological changes such as early cancer and sclerosing ulcers, and have important clinical diagnostic value and treatment guidance significance. These lesions are often key findings for diagnosis and require key attention and detailed evaluation. Therefore, images containing high-rigidity lesions should be assigned a high initial score to ensure they are ranked high in the final value ranking. Although low-rigidity tissues show abnormal texture in stable images, their motion characteristics in peristaltic images indicate normal elasticity, and may only be local variations or mild inflammation of normal tissue. The clinical significance of these areas is relatively limited, but they still have some reference value and can help understand the overall state of the digestive tract. Therefore, images containing low-rigidity tissues should be assigned a medium initial score and be considered as secondary focus. Images that do not contain lesion areas do not provide lesion information, but they still have some reference value and can be used as a control for normal tissues. Images containing only motion artifacts have relatively low value because the artifacts have no diagnostic significance. Therefore, images that do not contain lesion areas and images that contain only motion artifacts should be assigned a low initial score and be ranked low in the final ranking.

[0096] Secondly, the initial value score is adjusted based on the anatomical location labels. Specifically, a location weight coefficient is assigned to each anatomical location label, and the product of the initial value score and the corresponding location weight coefficient is calculated to obtain the value score after location adjustment. The location weight coefficient is determined based on medical statistics and clinical experience, reflecting the incidence of lesions and clinical importance of that anatomical location. For example, the gastric antrum, pyloric region, duodenal bulb, and ileocecal junction are high-incidence sites of digestive tract lesions and have higher diagnostic value, so they should be assigned a larger location weight coefficient. On the other hand, the location weight coefficient is relatively smaller for sites with relatively fewer lesions, such as the mid-esophagus and mid-small intestine.

[0097] Furthermore, the temporal distance between each image and the corresponding anatomical inflection point is calculated, and the value score is adjusted a second time based on the temporal distance. The temporal distance is the absolute value of the difference between the temporal index of the image in the filtered digestive tract image sequence and the temporal index of the nearest anatomical inflection point. The smaller the temporal distance, the closer the image is to the inflection point; the larger the temporal distance, the farther the image is from the inflection point.

[0098] The value score is adjusted based on temporal distance. Specifically, when the temporal distance is small, additional points are awarded; when the temporal distance is large, no additional adjustment is made or a small deduction is given. The amount of the bonus increases as the temporal distance decreases, ensuring that the image closest to the inflection point receives the highest bonus.

[0099] It should be noted that anatomical inflection points are the junctions of different anatomical regions of the digestive tract, which have special significance in clinical diagnosis. Images near inflection points often contain key anatomical landmarks, which help to accurately locate the anatomical location of lesions and formulate precise treatment plans. Therefore, images near anatomical inflection points have additional localization reference value.

[0100] In one alternative embodiment, the temporal distance adjustment can take the form of a decay function, such as exponential decay or linear decay. An exponential decay function can make images very close to the inflection point gain a bonus, while the bonus of images slightly farther away from the inflection point decays rapidly to zero. A linear decay function makes the bonus decrease linearly with distance, and the decay rate is relatively slow.

[0101] Furthermore, by combining the initial value score, location weight coefficient adjustment, and temporal distance adjustment, a comprehensive value score is obtained for each image. The selected digestive tract image sequences are then sorted in descending order according to the comprehensive value score, with the image with the highest score at the top and the image with the lowest score at the bottom. The sorted digestive tract image sequences are output as the value analysis results for viewing and diagnosis. In the output results, each image is accompanied by information such as lesion area annotation, lesion attribute label, anatomical location label, and comprehensive value score, which facilitates quick understanding of the key information of the image.

[0102] It should be noted that by quantifying and sorting the value of massive amounts of gastrointestinal images, the workload of image reading is reduced and diagnostic efficiency is improved. At the same time, because high-value images concentrate key lesion information and anatomical landmarks, important lesions can be identified and located in a shorter time, avoiding the omission of important lesions due to the large number of images. Secondly, the output of the comprehensive value score also provides a quantitative reference, which can more accurately reflect the comprehensive diagnostic value of the images, helping to allocate time and energy reasonably in the diagnostic process and optimize the diagnostic process.

[0103] On the other hand, this embodiment also provides an image value analysis system for a large number of images of the digestive tract, which includes:

[0104] The image classification module acquires a sequence of digestive tract images and calculates the optical flow field between adjacent frames. Based on the curl and divergence of the optical flow field, the digestive tract image sequence is classified to obtain motion type labels for the digestive tract image sequence. The motion type labels include stable images, peristaltic images, and non-peristaltic images.

[0105] The redundancy processing module uses the optical flow field to perform redundancy processing on the digestive tract image sequence to obtain a filtered digestive tract image sequence.

[0106] The anatomical location module extracts features from the filtered digestive tract image sequences, determines the anatomical inflection points of the filtered digestive tract image sequences, and labels the anatomical locations.

[0107] The lesion identification module performs image value analysis on the screened digestive tract image sequence based on the motion type label to identify lesion areas. The image value analysis includes analyzing the stable images to identify candidate lesion areas, combining the peristaltic images to perform motion characteristic analysis on the candidate lesion areas, and determining the lesion attributes of the candidate lesion areas based on the results of the motion characteristic analysis.

[0108] The image value module scores the value of the selected digestive tract image sequences based on the lesion area and the anatomical location label to obtain the value analysis results.

[0109] If the above functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0110] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-including system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0111] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0112] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0113] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for image value analysis of a large number of images of the digestive tract interior, characterized in that, include: A sequence of digestive tract images is acquired and the optical flow field between adjacent frames is calculated. The digestive tract image sequence is classified based on the curl and divergence of the optical flow field to obtain motion type labels for the digestive tract image sequence. The motion type labels include stable images, peristaltic images, and non-peristaltic images. The optical flow field is used to perform redundancy processing on the digestive tract image sequence to obtain a filtered digestive tract image sequence. Feature extraction is performed on the filtered digestive tract image sequence to determine the anatomical inflection points of the filtered digestive tract image sequence and label the anatomical locations. Image value analysis is performed on the screened digestive tract image sequences based on the motion type labels to identify lesion areas; the image value analysis includes analyzing the stable images to identify candidate lesion areas, combining the peristaltic images to perform motion characteristic analysis on the candidate lesion areas, and determining the lesion attributes of the candidate lesion areas based on the results of the motion characteristic analysis. The value of the selected digestive tract image sequences is scored based on the lesion area and the anatomical location label to obtain the value analysis results.

2. The image value analysis method for a large number of images of the digestive tract as described in claim 1, characterized in that: Based on the curl and divergence of the optical flow field, the digestive tract image sequence is classified to obtain motion type labels for the digestive tract image sequence. The divergence field and curl field are generated by calculating based on the horizontal and vertical components of the optical flow field. The divergence field and curl field are globally spatially aggregated to calculate global motion features that characterize the overall motion amplitude of the image sequence. The global motion features are then used to preliminarily divide the digestive tract image sequence into stable images and moving images. Pixel identification is performed on the divergence field corresponding to the moving image, and the identified pixels are clustered to obtain the divergence salient region; The optical flow vectors within the significant divergence region are extracted, and the directional distribution of the optical flow vectors is analyzed to obtain the directional consistency features of the optical flow vectors. Based on the directional consistency features, the motion image is further divided to obtain the creeping image and the non-creeping image, thus completing the labeling of the motion type.

3. The image value analysis method for a large number of images of the digestive tract as described in claim 2, characterized in that: The process of performing redundancy processing on the digestive tract image sequence using the optical flow field to obtain a filtered digestive tract image sequence includes: using the optical flow vector in the optical flow field to perform reverse mapping on the current frame to obtain a motion-compensated aligned frame; and calculating the pixel difference between the aligned frame and the previous frame. The corresponding redundancy threshold is obtained based on the motion type label of the current frame; among them, the first redundancy threshold is used for stable images, the second redundancy threshold is used for creeping images, and no redundancy threshold is set for non-creeping images, and all frames are directly removed; The pixel difference is compared with the corresponding redundancy threshold. When the pixel difference is less than or equal to the redundancy threshold, the current frame is determined to be a redundant frame and is removed. When the pixel difference is greater than the redundancy threshold, the current frame is retained, and the filtered digestive tract image sequence is obtained.

4. The image value analysis method for a large number of images of the digestive tract as described in claim 3, characterized in that: Feature extraction is performed on the filtered digestive tract image sequence to determine the anatomical inflection points of the filtered digestive tract image sequence and label the anatomical locations. This includes extracting color features, texture features, and edge features from each image in the filtered digestive tract image sequence and combining them to form a feature vector. The feature vector is reduced in dimensionality and mapped to the feature space to form a sequence of feature points. The Euclidean distance between adjacent feature points is then calculated. The locations of abrupt changes in the Euclidean distance are detected to obtain abrupt change points, and the images corresponding to the abrupt change points are marked as anatomical structural turning points. Based on the location of the inflection points of the anatomical structures, the image segments between adjacent inflection points are labeled with the corresponding anatomical location tags.

5. The image value analysis method for a large number of images of the digestive tract as described in claim 4, characterized in that: Analyzing the stable image to identify candidate lesion regions includes using a sliding window to traverse the filtered digestive tract image sequence corresponding to the stable image, calculating the information entropy of the pixel grayscale distribution within the sliding window and using it as the texture entropy value. For the sliding window, the degree of deviation of the texture entropy value of the sliding window relative to the texture entropy of the neighborhood is calculated, and the position where the deviation exceeds the judgment criterion is marked as the entropy mutation point; The entropy mutation markers at each image location are statistically analyzed, and the locations marked as entropy mutations are identified as candidate lesion regions.

6. The image value analysis method for a large number of images of the digestive tract as described in claim 5, characterized in that: Combining the peristaltic images to perform motion characteristic analysis on the candidate lesion regions, and determining the lesion attributes of the candidate lesion regions based on the results of the motion characteristic analysis, includes tracing the candidate lesion regions in the filtered digestive tract image sequence corresponding to the peristaltic images; The optical flow field is used to extract the local optical flow vector of the candidate lesion region and the background optical flow vector of the neighboring background region; The motion difference between the local optical flow vector and the background optical flow vector is calculated, and the lesion attributes of the candidate lesion region are determined based on the motion difference metric; the lesion attributes include high-rigidity lesions, low-rigidity tissues, and motion artifacts.

7. The image value analysis method for a large number of images of the digestive tract as described in claim 6, characterized in that: The value analysis results are obtained by scoring the selected digestive tract image sequences based on the lesion area and the anatomical location label. This includes assigning an initial value score to each image in the selected digestive tract image sequence based on the lesion area; wherein images containing high-rigidity lesions are assigned a high initial score, images containing low-rigidity tissues are assigned a medium initial score, and images that do not contain lesion areas or only contain motion artifacts are assigned a low initial score. Based on the anatomical location labels, the temporal distance between each image and the corresponding anatomical structure inflection point is calculated, and the initial value score is adjusted according to the temporal distance to obtain the comprehensive value score of each image. The selected digestive tract image sequences are arranged according to the comprehensive value score, and the sorted digestive tract image sequences are used as the value analysis results.

8. An image value analysis system for a large number of images of the digestive tract using the method described in any one of claims 1-7, characterized in that: The image classification module acquires a sequence of digestive tract images and calculates the optical flow field between adjacent frames. Based on the curl and divergence of the optical flow field, it classifies the digestive tract image sequence to obtain motion type labels for the digestive tract image sequence. The motion type labels include stable images, peristaltic images, and non-peristaltic images. The redundancy processing module uses the optical flow field to perform redundancy processing on the digestive tract image sequence to obtain a filtered digestive tract image sequence. The anatomical location module extracts features from the filtered digestive tract image sequence, determines the anatomical inflection points of the filtered digestive tract image sequence, and labels the anatomical locations. The lesion identification module performs image value analysis on the screened digestive tract image sequence based on the motion type label to identify lesion areas; the image value analysis includes analyzing the stable images to identify candidate lesion areas, combining the peristaltic images to perform motion characteristic analysis on the candidate lesion areas, and determining the lesion attributes of the candidate lesion areas based on the results of the motion characteristic analysis. The image value module scores the value of the selected digestive tract image sequences based on the lesion area and the anatomical location label to obtain the value analysis results.

9. A computer device, comprising: A memory and a processor; the memory stores a computer program, characterized in that: when the processor executes the computer program, it implements the steps of the method as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1-7.

Images