A tumor rehabilitation diagnosis and treatment system and method based on artificial intelligence

Abstract

The present application provides an artificial intelligence-based tumor rehabilitation diagnosis and treatment system and method. The system uses a global contour of the tumor edge in the target patient's tumor ultrasound image, and constructs an edge guidance baseline for fine-grained identification of the local tumor edge in the tumor ultrasound image based on the geometric constraint characteristics of the global contour edge. Multiple identification partitions of the local tumor edge are extracted from the tumor ultrasound image based on the projection intersection relationship between the edge guidance baseline and the global contour in the same projection space. The multi-dimensional target features of the tumor edge in the tumor ultrasound image are determined based on the edge wrinkle coefficient of the tumor nodule in each identification partition and the spatial difference relationship between the edge guidance baseline in each identification partition and the contour line of the corresponding global contour. The benign and malignant identification results of the target patient's thyroid tumor are determined based on the multi-dimensional target features. The solution provided by the present application can capture the invasive growth pattern of the tumor from the tumor ultrasound image.

Description

Technical Field

[0001] The present application relates to the technical field of medical image analysis, and more specifically, to an artificial intelligence-based tumor rehabilitation diagnosis and treatment system and method. Background Art

[0002] Medical image analysis is a technology that uses computer science and digital information processing technology to process, analyze and interpret medical imaging data. It has developed rapidly with the advancement of computer technology. The massive amount of medical image data poses a serious challenge to traditional manual interpretation methods. In recent years, with the development of artificial intelligence, medical image analysis has made major breakthroughs and is playing an increasingly critical role in disease diagnosis, treatment plan planning, and other aspects.

[0003] In existing medical image analysis, medical image analysis is mainly based on image processing and pattern recognition technology to pre-process medical images, extract feature information of the region of interest from the pre-processed medical images, and finally use machine learning to identify and analyze the extracted feature information, thereby providing doctors with accurate diagnostic basis; however, in tumor ultrasound image analysis in tumor rehabilitation diagnosis and treatment, the tumor edge in the tumor ultrasound image often shows weak edge characteristics. Existing technologies usually directly perform edge detection on the entire tumor ultrasound image, lacking the multi-dimensional quantification capability of the fine-grained morphological features of the tumor edge (such as wrinkling and spatial offset), and it is difficult to capture the aggressive growth pattern of the tumor (such as tumor edge wrinkling). Therefore, how to capture the aggressive growth pattern of the tumor from the tumor ultrasound image has become a difficult problem faced by the industry. Summary of the Invention

[0004] The present application provides an artificial intelligence-based tumor rehabilitation diagnosis and treatment system and method, which can capture the aggressive growth pattern of the tumor from tumor ultrasound images.

[0005] In a first aspect, the present application provides a method for intelligently identifying tumor ultrasound images, which is used in a tumor rehabilitation diagnosis and treatment system to perform malignant risk grading of thyroid tumors in target patients. The method comprises the following steps:

[0006] Triggering the tumor image recognition operation to obtain the tumor ultrasound image of the target patient during thyroid tumor rehabilitation diagnosis and treatment;

[0007] Performing global extraction of the tumor edge on the tumor ultrasound image to obtain a global contour of the tumor edge, and constructing an edge guidance baseline for fine-grained recognition of the local tumor edge in the tumor ultrasound image based on geometric constraint features of the global contour edge;

[0008] extracting a plurality of identification partitions of the local tumor edge in a fine-grained identification process from the tumor ultrasound image according to a projection intersection relationship between the edge guiding baseline and the global contour in the same projection space, and then determining an edge wrinkle coefficient of the tumor nodule in each identification partition;

[0009] The multidimensional target features of the tumor edge in the tumor ultrasound image are determined based on all edge wrinkle coefficients combined with the spatial difference relationship between the edge guide baseline and the contour line of the corresponding global contour in each identification partition, and the malignancy risk level of the target patient's thyroid tumor is identified based on the multidimensional target features.

[0010] In some embodiments, performing global extraction of the tumor edge on the tumor ultrasound image to obtain a global contour of the tumor edge specifically includes:

[0011] performing contrast enhancement on the tumor ultrasound image to obtain a tumor ultrasound enhanced image;

[0012] Extract all candidate edges from the tumor ultrasound enhancement image;

[0013] The region growing algorithm is used to connect all candidate edges to obtain the global outline of the tumor edge.

[0014] In some embodiments, constructing an edge guidance baseline for fine-grained recognition of a local tumor edge in the tumor ultrasound image based on the geometric constraint feature of the global contour edge specifically includes:

[0015] Constructing a minimum bounding rectangle of the global contour edge;

[0016] Extracting the geometric center point, maximum side length, and minimum side length of the minimum circumscribed rectangle;

[0017] Determining geometric constraint features of the global contour edge based on the geometric center point, the maximum side length, and the minimum side length;

[0018] The elliptic curve constructed by the geometric constraint feature is used as an edge guidance reference line when performing fine-grained recognition of the local tumor edge in the tumor ultrasound image.

[0019] In some embodiments, extracting multiple identification partitions of the local tumor edge in the fine-grained identification process from the tumor ultrasound image based on the projection intersection relationship between the edge guiding reference line and the global contour in the same projection space specifically includes:

[0020] Synchronously mapping the edge guidance reference line and the global contour to the same projection space, where the projection space is a representation of the tumor ultrasound image on a two-dimensional plane;

[0021] Extracting all intersection nodes between the edge guide reference line and the global contour in the same projection space, and describing the projection intersection relationship between the edge guide reference line and the global contour in the same projection space through each intersection node;

[0022] Multiple identification partitions are determined based on all adjacent intersection nodes when performing local feature recognition on the tumor edge.

[0023] In some embodiments, determining the edge wrinkle coefficient of the tumor nodule in each identified partition specifically includes:

[0024] Selecting an identification partition as a selected identification partition, and determining multiple curvature change rates and direction offsets on the edge contour of the tumor nodule in the selected identification partition;

[0025] A set of morphological change vectors of tumor nodules in the selected identification partition is constructed based on all curvature change rates and direction offsets;

[0026] Performing statistical analysis on the morphological change vector set to obtain edge wrinkle coefficients of tumor nodules within the identified partitions;

[0027] Continue to determine the margin fold coefficients of the tumor nodules within the remaining identified partitions.

[0028] In some embodiments, determining the multi-dimensional target features of the tumor edge in the tumor ultrasound image based on all edge wrinkle coefficients combined with the spatial difference relationship between the edge guide reference line in each identification partition and the contour line of the corresponding global contour specifically includes:

[0029] Obtain the edge wrinkle coefficient corresponding to each identified partition;

[0030] Calculating a local spatial offset feature between the edge guide reference line in each identification partition and the contour line of the corresponding global contour, and using the local spatial offset feature to describe a spatial difference relationship between the edge guide reference line in the identification partition and the contour line of the corresponding global contour;

[0031] Determine multiple tumor edge fusion indices based on the edge fold coefficient and spatial offset characteristics corresponding to each identified partition;

[0032] Multi-dimensional target features of the tumor edge in the tumor ultrasound image are extracted based on all tumor edge fusion indices.

[0033] In some embodiments, before performing global extraction of the tumor edge on the tumor ultrasound image, the method further includes: performing filtering processing on the tumor ultrasound image using Gaussian filtering.

[0034] In some embodiments, an ultrasound sensor is used to obtain ultrasound images of the tumor of the target patient during rehabilitation diagnosis and treatment of the thyroid tumor.

[0035] In some embodiments, the ultrasonic sensor is a Doppler ultrasonic sensor.

[0036] In a second aspect, the present application provides an artificial intelligence-based tumor rehabilitation diagnosis and treatment system, which includes a tumor ultrasound image intelligent recognition unit, and the tumor ultrasound image intelligent recognition unit includes:

[0037] An acquisition module is used to trigger the tumor image recognition operation and obtain the tumor ultrasound image of the target patient during thyroid tumor rehabilitation diagnosis and treatment;

[0038] a processing module, configured to perform global extraction of the tumor edge on the tumor ultrasound image to obtain a global contour of the tumor edge, and construct an edge guidance baseline for fine-grained identification of the local tumor edge in the tumor ultrasound image based on geometric constraint features of the global contour edge;

[0039] The processing module is further configured to extract a plurality of identification partitions of the local tumor edge in a fine-grained identification process from the tumor ultrasound image based on a projection intersection relationship between the edge guiding baseline and the global contour in the same projection space, and then determine an edge wrinkle coefficient of the tumor nodule in each identification partition;

[0040] An execution module is used to determine the multidimensional target features of the tumor edge in the tumor ultrasound image based on all edge wrinkle coefficients combined with the spatial difference relationship between the edge guide baseline and the contour line of the corresponding global contour in each identification partition, and identify the malignancy risk level of the target patient's thyroid tumor based on the multidimensional target features.

[0041] The technical solutions provided by the embodiments disclosed in this application have the following beneficial effects:

[0042] In the artificial intelligence-based tumor rehabilitation diagnosis and treatment system and method provided in the present application, first, a tumor ultrasound image of a target patient undergoing thyroid tumor rehabilitation diagnosis and treatment is obtained; secondly, a global extraction of the tumor edge is performed on the tumor ultrasound image to obtain a global contour of the tumor edge, and an edge guiding baseline is constructed for fine-grained identification of the local tumor edge in the tumor ultrasound image based on the geometric constraint characteristics of the global contour edge; then, multiple identification partitions of the local tumor edge in the fine-grained identification process are extracted from the tumor ultrasound image based on the projection intersection relationship between the edge guiding baseline and the global contour in the same projection space, and the edge fold coefficient of the tumor nodule in each identification partition is determined; finally, the multidimensional target features of the tumor edge in the tumor ultrasound image are determined based on all the edge fold coefficients combined with the spatial difference relationship between the edge guiding baseline in each identification partition and the contour line of the corresponding global contour, and the malignancy risk level of the target patient's thyroid tumor is identified based on the multidimensional target features.

[0043] It can be seen that the present application can capture the invasive growth pattern of the tumor from the tumor ultrasound image; first, obtain the tumor ultrasound image of the target patient during the rehabilitation diagnosis and treatment of thyroid tumor, and then provide a data basis for the target patient's thyroid benign and malignant identification; second, extract the global contour of the tumor edge in the tumor ultrasound image, and construct the edge guidance baseline for fine-grained identification of the local tumor edge in the tumor ultrasound image based on the geometric constraint characteristics of the global contour edge, so as to effectively establish a structured recognition reference system in the local image area with complex tumor edges, and accurately guide the fine-grained extraction of the tumor edge in the local range; further, extract the local tumor edge from the tumor ultrasound image in the fine-grained manner according to the projection intersection relationship between the edge guidance baseline and the global contour in the same projection space. Multiple recognition partitions are used in the recognition process to effectively limit the spatial range of local feature extraction, so that the subsequently extracted morphological features are more targeted at specific small-scale tumor edge changes, thereby improving the accuracy and precision of overall tumor recognition, and extracting the edge wrinkle coefficient of each recognition partition in the tumor ultrasound image to quantify the fine-grained morphological features of the tumor edge; then, based on all the edge wrinkle coefficients combined with the spatial difference relationship between the edge guide baseline in each recognition partition and the contour line of the corresponding global contour, the multi-dimensional target features of the tumor edge in the tumor ultrasound image are determined to perform multi-dimensional quantification of the fine-grained morphological features of the tumor edge, and depict the overall and local morphological change characteristics of the tumor edge from multiple angles and in a fine-grained manner, thereby capturing the invasive growth pattern of the tumor in the tumor ultrasound image. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 is an exemplary flow chart of a method for intelligently identifying tumor ultrasound images according to some embodiments of the present application;

[0045] Figure 2 This is a schematic diagram of an application scenario architecture of the method for intelligent tumor ultrasound image recognition according to some embodiments of the present application;

[0046] Figure 3 is an exemplary flow chart of determining a global profile according to some embodiments of the present application;

[0047] Figure 4 is an exemplary flow chart of determining and identifying partitions according to some embodiments of the present application;

[0048] Figure 5 is a schematic structural diagram of a tumor ultrasound image intelligent recognition unit according to some embodiments of the present application;

[0049] Figure 6 It is a structural diagram of a computer device for implementing an intelligent tumor ultrasound image recognition method according to some embodiments of the present application. DETAILED DESCRIPTION

[0050] In order to better understand the technical solution of the present application, the technical solution of the present application will be described in detail below with reference to the accompanying drawings and specific implementation methods.

[0051] refer to Figure 1 , which is an exemplary flow chart of a method for intelligently identifying tumor ultrasound images according to some embodiments of the present application. The method 100 for intelligently identifying tumor ultrasound images mainly includes the following steps:

[0052] In step 101, a tumor image recognition operation is triggered to obtain a tumor ultrasound image of a target patient undergoing thyroid tumor rehabilitation diagnosis and treatment.

[0053] In a specific implementation, the tumor image recognition operation is triggered, and an ultrasound sensor can be used to obtain a tumor ultrasound image of the target patient during thyroid tumor rehabilitation diagnosis and treatment. The ultrasound sensor can be a Doppler ultrasound sensor. In addition, other ultrasound sensors can be used to obtain a tumor ultrasound image of the target patient during thyroid tumor rehabilitation diagnosis and treatment, which is not limited here. Figure 2 This is a schematic diagram of the application scenario architecture of the intelligent tumor ultrasound image recognition method shown in some embodiments of the present application. The acquisition sensor interacts with the server through a communication network to complete the acquisition of image data. That is, the present application uses a Doppler ultrasound sensor to interact with the server through a communication network to complete the acquisition of tumor ultrasound images during thyroid tumor rehabilitation diagnosis and treatment of target patients, and decodes the received tumor ultrasound images through the server to store them in the data storage system for subsequent recognition, analysis and processing.

[0054] It should be noted that the tumor ultrasound image in this application represents a two-dimensional image generated after scanning the thyroid area of ​​the target patient. By determining the tumor ultrasound image, a data basis can be provided for identifying benign and malignant thyroid glands of the target patient.

[0055] In step 102, a global extraction of the tumor edge is performed on the tumor ultrasound image to obtain a global contour of the tumor edge, and an edge guidance baseline for fine-grained recognition of the local tumor edge in the tumor ultrasound image is constructed based on the geometric constraint features of the global contour edge.

[0056] It should be noted that the global contour in this application is the preliminary extraction result of the tumor boundary in the entire tumor ultrasound image. The global contour is a roughly extracted tumor shape boundary, which shows the basic position, shape and spatial layout of the tumor in the entire ultrasound image. By determining the global contour, the recognition range can be limited for subsequent fine-grained recognition of the mechanical energy tumor edge to avoid mistaking surrounding normal tissue or noise for tumors.

[0057] In a specific implementation, before performing global extraction of the tumor edge on the tumor ultrasound image, the process further includes: performing filtering processing on the tumor ultrasound image using Gaussian filtering, which will not be described in detail here.

[0058] In some embodiments, reference Figure 3 As shown in FIG. 1 , this figure is an exemplary flow chart of determining a global contour according to some embodiments of the present application. In this embodiment, a global extraction of the tumor edge is performed on the tumor ultrasound image to obtain a global contour of the tumor edge, which can be achieved by using the following steps:

[0059] First, in step 1021, contrast enhancement is performed on the tumor ultrasound image to obtain a tumor ultrasound enhanced image;

[0060] Then, in step 1022, all candidate edges in the tumor ultrasound enhancement image are extracted;

[0061] Finally, in step 1023, a region growing algorithm is used to connect all candidate edges to obtain a global outline of the tumor edge.

[0062] In a specific implementation, first, an adaptive contrast enhancement algorithm in image processing can be used to perform contrast enhancement on the tumor ultrasound image to obtain a tumor ultrasound-enhanced image. For example, the adaptive contrast enhancement algorithm can make the low-contrast tumor edge in the tumor ultrasound image clearer by contrast stretching a small local block, thereby completing the contrast enhancement of the tumor ultrasound image. In addition, histogram equalization can also be used to enhance the contrast of the tumor ultrasound image. Second, based on a gradient operator, the outermost edge is extracted from the tumor ultrasound-enhanced image as a candidate edge, and then all candidate edges are obtained. The gradient operator can be a Sobel operator or a Canny operator, which is not limited here. Finally, an existing region growing algorithm is used to connect all candidate edges, and the closed curve obtained by the connection is used as the global contour of the tumor edge.

[0063] It should be noted that, in this embodiment, the tumor ultrasound enhanced image represents the image after contrast enhancement of the tumor ultrasound image; in this embodiment, the candidate edge represents a specific edge extracted from the tumor ultrasound image. Specifically, the candidate edge is the outermost edge in the tumor ultrasound enhanced image. In the identification of thyroid tumors in target patients, the outermost boundary of the thyroid tumor area is usually the core feature for determining whether the tumor has worsened. Therefore, in this application, the outermost edge in the tumor ultrasound enhanced image can be used as a candidate edge to more effectively identify the deterioration of the thyroid tumor.

[0064] In some embodiments, constructing an edge-guiding baseline for fine-grained identification of a local tumor edge in the tumor ultrasound image based on the geometric constraint features of the global contour edge can be achieved by the following steps, namely:

[0065] Constructing a minimum bounding rectangle of the global contour edge;

[0066] Extracting the geometric center point, maximum side length, and minimum side length of the minimum circumscribed rectangle;

[0067] Determining geometric constraint features of the global contour edge based on the geometric center point, the maximum side length, and the minimum side length;

[0068] The elliptic curve constructed by the geometric constraint feature is used as an edge guidance reference line when performing fine-grained recognition of the local tumor edge in the tumor ultrasound image.

[0069] It should be noted that the edge guidance baseline in this application represents a reference line used to assist in identifying the local change characteristics of the tumor edge. By arranging the edge guidance baseline in the tumor ultrasound image, a structured recognition reference system can be effectively established in the complex local image area of ​​the tumor edge, and the fine-grained extraction of the tumor edge can be accurately guided within the local range, avoiding the problem of local feature extraction offset in an environment with a large influence of ultrasonic noise.

[0070] In the specific implementation, first, the existing rotating caliper method is used to construct the minimum circumscribed rectangle of the contour edge from the global contour, that is, the rotating caliper method simulates placing a set of "mutually perpendicular" calipers (straight line tools) on the global contour, and then rotates the calipers along the edge of the global contour to find the rectangle with the smallest area that can completely cover the global contour as the minimum circumscribed rectangle of the global contour edge; secondly, the geometric center point, maximum side length and minimum side length of the minimum circumscribed rectangle are extracted through the data processing tool Python; then, the geometric constraint features of the global contour edge are determined based on the geometric center point, the maximum side length and the minimum side length, that is: the geometric center point, the maximum side length and the minimum side length can be used as the minimum circumscribed rectangle of the global contour edge. The set of the center point, the maximum side length and the minimum side length is used as the geometric constraint feature of the global contour edge; finally, the elliptic curve constructed by the geometric constraint feature is used as the edge guidance baseline for fine-grained recognition of the local tumor edge in the tumor ultrasound image, that is: the geometric center point, the maximum side length and the minimum side length corresponding to the geometric constraint feature are obtained, the geometric center point is used as the center position of the ellipse, the maximum side length and the minimum side length are used as the major axis and minor axis lengths of the ellipse respectively, an elliptic curve centered on the geometric center point is constructed, and the elliptic curve is used as the edge guidance baseline for fine-grained recognition of the local tumor edge in the tumor ultrasound image.

[0071] It should be noted that the geometric constraint feature in this embodiment represents the feature used to limit the overall spatial distribution range of the tumor contour. The determination of the geometric constraint feature ensures that the auxiliary line not only covers the overall shape but also can adapt to the local recognition needs; in this embodiment, the elliptical curve represents a regular closed curve that depicts an ellipse.

[0072] In step 103, multiple identification partitions of the local tumor edge in the fine-grained identification process are extracted from the tumor ultrasound image based on the projection intersection relationship between the edge guidance baseline and the global contour in the same projection space, and then the edge wrinkle coefficient of the tumor nodule in each identification partition is determined.

[0073] It should be noted that the identification partition in this application represents the local area formed by the edge guiding baseline and the global contour of the tumor in the same projection space. The local area corresponds to a small range of the tumor edge and is used to extract local features separately in the fine-grained recognition process. The identification partition usually includes the global contour arc segment between two intersection nodes and the spatial sub-area surrounded by the corresponding auxiliary line arc segment. The identification partition can effectively limit the spatial range of local feature extraction, so that the subsequently extracted morphological features are more targeted at specific small-scale tumor edge changes, thereby improving the accuracy and precision of overall tumor recognition.

[0074] In some embodiments, reference Figure 4 As shown in FIG. 1 , this figure is an exemplary flow chart for determining identification partitions according to some embodiments of the present application. In this embodiment, extracting multiple identification partitions of the local tumor edge from the tumor ultrasound image in the fine-grained identification process based on the projection intersection relationship between the edge guiding baseline and the global contour in the same projection space can be achieved using the following steps:

[0075] First, in step 1031, the edge guidance reference line and the global contour are synchronously mapped to the same projection space, where the projection space is a representation of the tumor ultrasound image on a two-dimensional plane;

[0076] Then, in step 1032, all intersection nodes of the edge guide reference line and the global contour in the same projection space are extracted, and the projection intersection relationship between the edge guide reference line and the global contour in the same projection space is described by each intersection node;

[0077] Finally, in step 1033 , multiple identification partitions for performing local feature recognition on the tumor edge are determined based on all adjacent intersection nodes.

[0078] In a specific implementation, first, the edge guiding reference line and the global contour are synchronously mapped to the same projection space, where the projection space is a representation of the tumor ultrasound image on a two-dimensional plane. That is, the edge guiding reference line and the global contour are mapped to the pixel coordinates of the tumor ultrasound image, thereby completing the synchronous mapping of the edge guiding reference line and the global contour. Then, using the image processing tool OpenCV, all intersection nodes of the edge guiding reference line and the global contour in the same projection space are extracted. That is, the intersection point between the intersection node edge guiding reference line and the global contour is regarded as the intersection node, and each intersection node is used to describe the projection intersection relationship between the edge guiding reference line and the global contour in the same projection space. Finally, based on all adjacent intersection nodes, multiple identification partitions are determined for local feature recognition of the tumor edge. That is, the closed area formed by every two adjacent intersection nodes is used as the identification partition for local feature recognition of the tumor edge, thereby obtaining multiple identification partitions for local feature recognition of the tumor edge. The identification partition is composed of the closed area between the edge guiding reference line and the contour line of the global contour.

[0079] It should be noted that, in this embodiment, the intersection node represents the intersection point formed by the intersection of the edge guide reference line and the contour line of the global contour on the image plane, and the intersection node is the basic anchor point for subsequent local recognition partition division.

[0080] In some embodiments, determining the edge wrinkle coefficient of the tumor nodule in each identified partition may be achieved by the following steps:

[0081] Selecting an identification partition as a selected identification partition, and determining multiple curvature change rates and direction offsets on the edge contour of the tumor nodule in the selected identification partition;

[0082] A set of morphological change vectors of tumor nodules in the selected identification partition is constructed based on all curvature change rates and direction offsets;

[0083] Performing statistical analysis on the morphological change vector set to obtain edge wrinkle coefficients of tumor nodules within the identified partitions;

[0084] Continue to determine the margin fold coefficients of the tumor nodules within the remaining identified partitions.

[0085] In this application, the edge wrinkle coefficient represents an indicator that quantifies the degree of wrinkling of the edge of a tumor nodule, and is used to describe the variability of the concave and convex sides of the edge of the tumor nodule, thereby reflecting the abnormality and invasiveness of tumor growth. In actual analysis, the higher the edge wrinkle coefficient, the more complex the edge of the tumor nodule in the local area, indicating a greater potential risk of malignancy. Therefore, by determining the edge wrinkle coefficient, the benign and malignant nature of the tumor can be identified in a fine-grained manner.

[0086] In the specific implementation, first, multiple curvature change rates and direction offsets on the edge contour of the tumor nodule in the selected identification partition are determined, that is, multiple sampling points on the edge contour of the tumor nodule in the selected identification partition are obtained in an equally spaced sampling manner, and the change rate of the local tangent direction formed by each adjacent sampling point is used as the curvature change rate on the edge contour of the tumor nodule in the selected identification partition, and the angle between the direction of the connecting line between each adjacent sampling point and the average direction of the global contour is used as the offset angle on the edge contour of the tumor nodule in the selected identification partition, wherein the corresponding global contour line in the identification partition can be used as the edge contour of the tumor nodule; secondly, a morphological change vector set of the tumor nodule in the selected identification partition is constructed based on all the curvature change rates and direction offsets, that is, the curvature change rate and direction offset corresponding to each adjacent sampling point are combined into a two-dimensional morphological change vector to obtain the selected identification partition in the tumor super A set of morphological change vectors in the acoustic image, wherein the morphological change vector set is composed of multiple morphological change vectors; then, the morphological change vector set is statistically analyzed to obtain the edge fold coefficient of the tumor nodule in the identification partition, that is, the information entropy of all morphological change vectors in the morphological change vector set can be used as the edge fold coefficient of the tumor nodule in the identification partition. In thyroid tumor identification, when a malignant tumor edge appears, the tumor nodule edge usually presents a higher curvature mutation and directional disorder. By counting the statistical parameters of the local area (that is, the information entropy) and using the statistical parameters as the edge fold coefficient of the tumor nodule in the identification partition, the morphological differences between benign and malignant can be effectively distinguished; finally, the edge fold coefficients of the tumor nodules in the remaining identification partitions are further determined by the determination method of "performing a statistical analysis on the morphological change vector set to obtain the edge fold coefficient of the tumor nodule in the identification partition".

[0087] It should be noted that in this embodiment, the curvature change rate represents the rate at which the contour curvature value between adjacent sampling points on the local tumor edge contour changes with the contour arc length; the directional offset in this embodiment represents the degree of deviation between the actual tumor edge trend and the auxiliary geometric shape; and the morphological change vector in this embodiment represents a two-dimensional vector formed by the combination of the curvature change rate and the directional offset.

[0088] In step 104, the multidimensional target features of the tumor edge in the tumor ultrasound image are determined based on all edge wrinkle coefficients combined with the spatial difference relationship between the edge guide baseline in each identification partition and the contour line of the corresponding global contour, and the malignancy risk level of the target patient's thyroid tumor is identified based on the multidimensional target features.

[0089] It should be noted that the multidimensional target features in this application represent multiple feature areas with rich edges calibrated in the tumor ultrasound image. The multidimensional target features can quantify the geometric morphology and spatial relationship of the tumor edge in the tumor ultrasound image in a fine-grained manner. The multidimensional target features not only include the morphological complexity information of the tumor edge at the local scale, but also include the spatial position deviation information of the tumor edge relative to the edge guide baseline. By determining the multidimensional target features, the fine-grained morphological features of the tumor edge can be multi-dimensionally quantified, and the overall and local morphological change characteristics of the tumor edge can be depicted from multiple angles and in a fine-grained manner, thereby capturing the aggressive growth pattern of the tumor in the tumor ultrasound image.

[0090] In some embodiments, determining the multi-dimensional target features of the tumor edge in the tumor ultrasound image based on all edge wrinkle coefficients combined with the spatial difference relationship between the edge guide reference line in each identification partition and the contour line of the corresponding global contour can be achieved by the following steps, namely:

[0091] Obtain the edge wrinkle coefficient corresponding to each identified partition;

[0092] Calculating a local spatial offset feature between the edge guide reference line in each identification partition and the contour line of the corresponding global contour, and using the local spatial offset feature to describe a spatial difference relationship between the edge guide reference line in the identification partition and the contour line of the corresponding global contour;

[0093] Determine multiple tumor edge fusion indices based on the edge fold coefficient and spatial offset characteristics corresponding to each identified partition;

[0094] Multi-dimensional target features of the tumor edge in the tumor ultrasound image are extracted based on all tumor edge fusion indices.

[0095] In the specific implementation, first, the edge fold coefficient corresponding to each identification partition is obtained; secondly, for each identification partition, the Euclidean distance between the edge guide baseline in the identification partition and the contour line of the global contour at each sampling point can be calculated, and the variance of all Euclidean distances can be used as the local spatial offset feature between the edge guide baseline in the identification partition and the contour line of the corresponding global contour, thereby obtaining the local spatial offset feature between the edge guide baseline in each identification partition and the contour line of the corresponding global contour; then, based on the edge fold coefficient and spatial offset feature corresponding to each identification partition, multiple tumor edge fusion indices are determined, that is: for each identification partition, the edge fold coefficient and spatial offset feature corresponding to the identification partition are normalized, and the weighted sum of the normalized edge fold coefficient and spatial offset feature is used as the tumor edge fusion index corresponding to the identification partition, thereby obtaining multiple tumor edge fusion indices, among which the maximum and minimum normalization can be used. The edge fold coefficient and spatial offset feature corresponding to the identification partition are normalized to between 0 and 1, which will not be repeated here. In the weighted summation of the normalized edge fold coefficient and the spatial offset feature, the weights of the edge fold coefficient and the spatial offset feature can be set according to their contribution to the identification of benign and malignant tumor edges. For example, in this application, the weights of the edge fold coefficient and the spatial offset feature are set to 0.57 and 0.43, respectively, which are not limited here. Finally, each tumor edge fusion index is compared with the fusion index threshold, and the identification partition with a tumor edge fusion index greater than the fusion index threshold is extracted as the target feature of the tumor edge in the tumor ultrasound image, and all the extracted target features are combined into a multi-dimensional target feature of the tumor edge in the tumor ultrasound image. The fusion index threshold can be set by learning a large number of tumor edge fusion indices based on machine learning or by expert knowledge, which is not limited here.

[0096] It should be noted that the spatial difference relationship in the present application represents the offset relationship in space between the edge guide baseline in the identification partition and the contour line of the corresponding global contour, which is described by the local spatial offset feature, that is, the local spatial offset feature represents the parameter describing the degree of offset between the edge guide baseline in the identification partition and the corresponding global contour line; in this embodiment, the tumor edge fusion index represents the characteristic index obtained by fusing the edge wrinkle coefficient corresponding to the identification partition area with the local spatial offset feature. The tumor edge fusion index simultaneously includes information on local morphological changes and spatial geometric offset information, and can more comprehensively reflect the fine-grained structural characteristics and abnormality degree of the tumor edge in different local areas, thereby providing a more accurate and richer data basis for the subsequent construction of the overall feature surface and the task of identifying benign and malignant tumors.

[0097] In some embodiments, identifying the malignancy risk level of the target patient's thyroid tumor based on the multi-dimensional target features can be achieved by the following steps, namely:

[0098] Inputting the multidimensional target features into a trained tumor classification model;

[0099] The tumor classification model is used to perform feature hierarchical analysis and output the malignancy risk level of the target patient's thyroid tumor.

[0100] In specific implementation, first, the multidimensional target features are input into a trained tumor classification model, which is a support vector machine classifier. In addition, the tumor classification model can also be a convolutional neural network classifier. During the training phase, the tumor classification model has performed malignant risk level classification learning and feature discrimination optimization on a large number of thyroid tumor image samples labeled with benign and malignant labels, which will not be repeated here. Then, after inputting the multidimensional target features, the tumor classification model first performs feature hierarchical analysis on the multidimensional target features in its hidden layer, that is, through multiple nonlinear transformations and feature weighted combinations, it automatically extracts high-order feature expressions for distinguishing benign and malignant tumors, and then maps the high-order feature expressions to the malignant risk level of the thyroid tumor through an activation function (such as Softmax or Sigmoid) in the output layer, and finally outputs the malignant risk level.

[0101] It should be noted that the tumor classification model in this embodiment represents a model for classifying and identifying the malignancy risk level of the target patient's thyroid tumor; the malignancy risk level in this application represents an indicator of the malignancy degree of the target patient's thyroid tumor.

[0102] In addition, in another aspect of the present application, in some embodiments, the present application provides an artificial intelligence-based tumor rehabilitation diagnosis and treatment system, the system includes a tumor ultrasound image intelligent recognition unit, reference Figure 5 This figure is a schematic diagram of the structure of a tumor ultrasound image intelligent recognition unit according to some embodiments of the present application. The tumor ultrasound image intelligent recognition unit 200 includes: an acquisition module 201, a processing module 202, and an execution module 203, which are described as follows:

[0103] Acquisition module 201, in this application, acquisition module 201 is mainly used to trigger the tumor image recognition operation and obtain the tumor ultrasound image of the target patient during thyroid tumor rehabilitation diagnosis and treatment;

[0104] Processing module 202, in the present application, is mainly used to perform global extraction of the tumor edge in the tumor ultrasound image to obtain a global contour of the tumor edge, and construct an edge guidance baseline for fine-grained recognition of the local tumor edge in the tumor ultrasound image based on the geometric constraint characteristics of the global contour edge;

[0105] The processing module 202 is further configured to extract a plurality of identification partitions of the local tumor edge in the fine-grained identification process from the tumor ultrasound image based on the projection intersection relationship between the edge guiding baseline and the global contour in the same projection space, and then determine the edge wrinkle coefficient of the tumor nodule in each identification partition;

[0106] Execution module 203, in this application, the execution module 203 is mainly used to determine the multidimensional target features of the tumor edge in the tumor ultrasound image based on all edge wrinkle coefficients combined with the spatial difference relationship between the edge guide baseline in each identification partition and the contour line of the corresponding global contour, and identify the malignancy risk level of the target patient's thyroid tumor based on the multidimensional target features.

[0107] In addition, the present application also provides a computer device, which includes a memory and a processor, wherein the memory stores code, and the processor is configured to obtain the code and execute the above-mentioned tumor ultrasound image intelligent recognition method.

[0108] In some embodiments, reference Figure 6 , which is a schematic diagram of the structure of a computer device for implementing the intelligent tumor ultrasound image recognition method according to some embodiments of the present application. The intelligent tumor ultrasound image recognition method in the above embodiment can be Figure 6 The computer device 300 shown in FIG. 1 is implemented as shown in FIG. 1 , and the computer device 300 includes at least one processor 301 , a communication bus 302 , a memory 303 , and at least one communication interface 304 .

[0109] The processor 301 may be a general-purpose central processing unit (CPU), or an application-specific integrated circuit (ASIC) or one or more processors for controlling the execution of the intelligent tumor ultrasound image recognition method of the present application.

[0110] The communication bus 302 may be used to transmit information between the aforementioned components.

[0111] Memory 303 may be, but is not limited to, a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, a random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, an electrically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM) or other optical disc storage, an optical disc storage (including compact discs, laser discs, optical discs, digital versatile discs, Blu-ray discs, etc.), a magnetic disk or other magnetic storage device, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer. Memory 303 may be independent and connected to processor 301 via communication bus 302. Memory 303 may also be integrated with processor 301.

[0112] Memory 303 is used to store program code for executing the solution of the present application, and is controlled by processor 301 for execution. Processor 301 is used to execute the program code stored in memory 303. The program code may include one or more software modules. The determination of the intelligent tumor ultrasound image recognition method in the above embodiment can be implemented by processor 301 and one or more software modules in the program code in memory 303.

[0113] The communication interface 304 uses any device such as a transceiver to communicate with other devices or communication networks, such as Ethernet, radio access network (RAN), wireless local area network (WLAN), etc.

[0114] In a specific implementation, as an example, a computer device may include multiple processors, each of which may be a single-core (single-CPU) processor or a multi-core (multi-CPU) processor. A processor herein may refer to one or more devices, circuits, and / or processing cores for processing data (e.g., computer program instructions).

[0115] The aforementioned computer device can be a general-purpose computer device or a dedicated computer device. In a specific implementation, the computer device can be a desktop computer, a portable computer, a network server, a personal digital assistant (PDA), a mobile phone, a tablet computer, a wireless terminal device, a communication device, or an embedded device. The embodiments of this application do not limit the type of computer device.

[0116] In addition, the present application also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the above-mentioned intelligent tumor ultrasound image recognition method.

[0117] Although the preferred embodiments of the present application have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present application.

[0118] Obviously, those skilled in the art may make various changes and modifications to this application without departing from the spirit and scope of this application. Thus, if these modifications and variations of this application fall within the scope of the claims of this application and their equivalents, this application is intended to include these modifications and variations.

Claims

1. A method for intelligent tumor ultrasound image recognition, used in a tumor rehabilitation diagnosis and treatment system to grade the malignant risk of thyroid tumors in target patients, characterized by: The method comprises the following steps: Triggering the tumor image recognition operation to obtain the tumor ultrasound image of the target patient during thyroid tumor rehabilitation diagnosis and treatment; Performing global extraction of the tumor edge on the tumor ultrasound image to obtain a global contour of the tumor edge, and constructing an edge guidance baseline for fine-grained recognition of the local tumor edge in the tumor ultrasound image based on geometric constraint features of the global contour edge, specifically comprising: constructing a minimum circumscribed rectangle of the global contour edge; Extracting the geometric center point, maximum side length, and minimum side length of the minimum circumscribed rectangle; Determining geometric constraint features of the global contour edge based on the geometric center point, the maximum side length, and the minimum side length; using the elliptic curve constructed by the geometric constraint feature as an edge guidance reference line for fine-grained identification of a local tumor edge in the tumor ultrasound image, wherein the edge guidance reference line represents a reference line for assisting in identifying local variation features of the tumor edge; extracting a plurality of identification partitions of the local tumor edge in a fine-grained identification process from the tumor ultrasound image according to a projection intersection relationship between the edge guiding baseline and the global contour in the same projection space, and then determining an edge wrinkle coefficient of the tumor nodule in each identification partition; Determining the multidimensional target features of the tumor edge in the tumor ultrasound image based on all edge wrinkle coefficients combined with the spatial difference relationship between the edge guide baseline and the contour line of the corresponding global contour in each identification partition, and identifying the malignancy risk level of the target patient's thyroid tumor based on the multidimensional target features, specifically including: Obtain the edge wrinkle coefficient corresponding to each identified partition; Calculating the local spatial offset feature between the edge-guided reference line and the contour line of the corresponding global contour in each recognition partition; Determine multiple tumor edge fusion indices based on the edge fold coefficient and local spatial offset characteristics corresponding to each identified partition; The identified partitions whose tumor edge fusion index is greater than the fusion index threshold are used as target features of the tumor edge in the tumor ultrasound image, and all the extracted target features are combined into multi-dimensional target features of the tumor edge in the tumor ultrasound image.

2. The method according to claim 1, wherein Performing global extraction of the tumor edge on the tumor ultrasound image to obtain a global contour of the tumor edge specifically includes: performing contrast enhancement on the tumor ultrasound image to obtain a tumor ultrasound enhanced image; Extract all candidate edges from the tumor ultrasound enhancement image; The region growing algorithm is used to connect all candidate edges to obtain the global outline of the tumor edge.

3. The method according to claim 1, wherein Extracting multiple identification partitions of the local tumor edge in the fine-grained identification process from the tumor ultrasound image according to the projection intersection relationship between the edge guiding reference line and the global contour in the same projection space specifically includes: Synchronously mapping the edge guidance reference line and the global contour to the same projection space, where the projection space is a representation of the tumor ultrasound image on a two-dimensional plane; Extracting all intersection nodes between the edge guide reference line and the global contour in the same projection space, and describing the projection intersection relationship between the edge guide reference line and the global contour in the same projection space through each intersection node; Multiple identification partitions are determined based on all adjacent intersection nodes when performing local feature recognition on the tumor edge.

4. The method according to claim 1, wherein Determining the edge wrinkle coefficient of the tumor nodule in each identification zone specifically includes: Selecting an identification partition as a selected identification partition, and determining multiple curvature change rates and direction offsets on the edge contour of the tumor nodule in the selected identification partition; A set of morphological change vectors of tumor nodules in the selected identification partition is constructed based on all curvature change rates and direction offsets; Performing statistical analysis on the morphological change vector set to obtain edge wrinkle coefficients of tumor nodules within the identified partitions; Continue to determine the margin fold coefficients of the tumor nodules within the remaining identified partitions.

5. The method according to claim 1, wherein Before performing global extraction of the tumor edge on the tumor ultrasound image, the method further includes: performing filtering processing on the tumor ultrasound image using Gaussian filtering.

6. The method according to claim 1, wherein Ultrasonic images of tumors of target patients undergoing thyroid tumor rehabilitation diagnosis and treatment are obtained through ultrasonic sensors.

7. The method according to claim 6, wherein The ultrasonic sensor is a Doppler ultrasonic sensor.

8. An artificial intelligence-based tumor rehabilitation diagnosis and treatment system, which uses the method according to any one of claims 1 to 7 to perform intelligent tumor ultrasound image recognition, the system comprising a tumor ultrasound image intelligent recognition unit, characterized in that: The tumor ultrasound image intelligent recognition unit includes: An acquisition module is used to trigger the tumor image recognition operation and obtain the tumor ultrasound image of the target patient during thyroid tumor rehabilitation diagnosis and treatment; a processing module, configured to perform global extraction of the tumor edge on the tumor ultrasound image to obtain a global contour of the tumor edge, and construct an edge guidance baseline for fine-grained identification of the local tumor edge in the tumor ultrasound image based on geometric constraint features of the global contour edge; The processing module is further configured to extract a plurality of identification partitions of the local tumor edge in a fine-grained identification process from the tumor ultrasound image based on a projection intersection relationship between the edge guiding baseline and the global contour in the same projection space, and then determine an edge wrinkle coefficient of the tumor nodule in each identification partition; An execution module is used to determine the multidimensional target features of the tumor edge in the tumor ultrasound image based on all edge wrinkle coefficients combined with the spatial difference relationship between the edge guide baseline and the contour line of the corresponding global contour in each identification partition, and identify the malignancy risk level of the target patient's thyroid tumor based on the multidimensional target features.

Images