Classification method and system based on thyroid or breast nodule

Abstract

The application relates to the technical field of medical image analysis, in particular to a classification method and system based on thyroid or breast nodules, which comprises the following steps: acquiring a continuous image frame sequence to locate a calcification region mapping space coordinate, associating adjacent frame homologous segments, extracting a nodule outer edge feature corresponding to a final frame calcification position, analyzing a gray peak value direction connectivity, extracting a texture aggregation boundary feature and overlapping volume curvature and echo distribution analysis to obtain a classification result. In the application, the spatial position and time variation of the calcification region in the continuous image frame are jointly analyzed to obtain a dynamic migration track corresponding to the nodule outer edge feature, a propagation path is constructed in combination with the direction distribution of the internal gray peak value, the boundary contact feature and the texture aggregation form are matched in a unified space, a multi-dimensional feature set with spatial continuity and direction integrity is formed, the analysis ability of the nodule volume, curvature and internal structure difference is improved, and the stability and distinguishability of the classification result are enhanced.

Description

Technical Field

[0001] This invention relates to the field of medical image analysis technology, and in particular to a classification method and system based on thyroid or breast nodules. Background Technology

[0002] Medical image analysis technology is an important area of ​​intersection between medical information processing and artificial intelligence. Its core aspects include medical image acquisition, preprocessing, feature extraction, image recognition and classification, and assisted diagnosis. This overall technological field is based on computer vision and clinical medical knowledge, combining multimodal medical imaging data. Through specific image analysis methods, it performs quantitative and qualitative analysis of the structure or lesions of internal tissues and organs. Common applications include MRI image segmentation, CT image recognition, and ultrasound image interpretation, with wide-ranging value in disease-assisted diagnosis, efficacy evaluation, and preoperative planning. Traditional classification methods based on thyroid or breast nodules involve identifying key imaging features in ultrasound images, such as nodule morphology, edges, echo intensity, and calcification distribution. Then, according to established imaging standards such as TI-RADS or BI-RADS grading systems, nodules are categorized according to their benign or malignant risk. Common methods include image feature scoring based on human experience, statistical discrimination methods, and classifiers such as support vector machines or convolutional neural networks for image feature input and output mapping. The nodule category is determined by constructing and training a model through an image feature space.

[0003] In the process of nodule classification based on static images, existing technologies often only analyze features at a single moment, lacking a systematic characterization of dynamic changes and inter-frame relationships. When the boundary contact position or internal structure undergoes slight displacement with the detection environment, it is easy to cause discontinuity of local features or loss of spatial correlation. For example, under conditions of breathing fluctuations or probe fine-tuning, the distribution of key structures may appear broken or shifted in the image, resulting in unstable classification basis and affecting the judgment of consistency between morphological and texture features. Summary of the Invention

[0004] To address the technical problems existing in the prior art, embodiments of the present invention provide a classification method based on thyroid or breast nodules, comprising the following steps: To achieve the above objectives, the present invention adopts the following technical solution: a classification method based on thyroid or breast nodules, comprising the following steps: S1: Locate the calcified pixel region in the continuous image frame sequence, map it to the image space coordinates, and associate the homologous calcified segments in adjacent frames according to the continuous change of position over time to obtain the calcification migration mapping path. S2: Based on the calcification migration mapping path, extract the nodule outer edge contour points, identify bending and acute angle segments, and obtain the calcification edge contact distribution structure corresponding to the calcification location space of the last frame. S3: Based on the calcified edge contact distribution structure, divide equal-area gray-scale sub-regions and compare the gray-scale distribution of the sub-regions, locate key peaks, analyze the connectivity of adjacent peaks according to the direction, connect adjacent peaks, present a continuous extension path and overall extension shape, and obtain a gray-scale structure extension trajectory diagram. S4: Based on the grayscale structure extension trajectory map, analyze the correspondence between the concentrated area of ​​the extension direction and the outer edge of the nodule, extract the boundary range of the concentrated direction, draw the texture aggregation pattern, unify the spatial alignment boundary features, and obtain the nodule texture aggregation boundary feature set; S5: Based on the nodule texture aggregation boundary feature set, perform feature position overlap on the calcified edge contact distribution structure, analyze the relationship between nodule volume, external curvature and internal echo distribution, combine discrimination elements, and obtain nodule classification results.

[0005] As a further embodiment of the present invention, the calcification migration mapping path includes the initial position of the calcification region, the intermediate transition position of the calcification region, and the final position of the calcification region; the calcification edge contact distribution structure includes the edge contact point position, the spatial distribution of contact points, and the contact point density; the grayscale structure extension trajectory map includes the grayscale peak position distribution, the peak connection path shape, and the overall extension direction; the nodule texture aggregation boundary feature set includes the texture aggregation region shape, aggregation boundary line features, and spatial reference system alignment features; and the nodule classification result includes volume features, external curvature features, and internal echo distribution features.

[0006] As a further aspect of the present invention, the homologous calcification segment refers to comparing the centroid position of the calcified pixel region in a continuous image frame sequence according to the time sequence, and connecting the calcified regions with the same position and morphology in adjacent frames in sequence to continue the connection process. The nodule outer edge contour points refer to the set of outer edge coordinate points extracted sequentially along the nodule boundary in the corresponding image frame. Contour segments with bending or acute angle features are identified, and the contact between the calcified area and the nodule edge is analyzed.

[0007] As a further aspect of the present invention, the calcification location space refers to the coordinate mapping of the initial position, intermediate position and position of the calcification region in the image plane or three-dimensional space, and the spatial comparison with the coordinates of the outer edge of the nodule to analyze the contact distribution between the calcification region and the edge of the nodule in the differential stage. The texture aggregation morphology refers to finding densely distributed boundary segments in the nodule region based on the grayscale structure extension trajectory map, and sequentially connecting adjacent boundary segments to obtain the outer contour of the texture aggregation region under a unified reference system, reflecting the concentrated distribution state of the echo texture inside the nodule.

[0008] As a further aspect of the present invention, the specific steps of S1 are as follows: S101: Obtain a continuous image frame sequence of densely calcified areas in breast or thyroid nodule images, locate the image region of each frame, identify the distribution of calcified pixels, match the identified pixel regions with the spatial coordinates of the frames, and connect the spatial coordinates sequentially in time order to obtain the spatial sequence of calcified pixels. S102: Based on the calcified pixel spatial sequence, analyze the spatial coordinate position relationship of the calcified region in adjacent frames, determine the direction of movement of the position in the time sequence, connect the continuous spatial change path in the time continuity manner, and obtain the calcified position change trajectory. S103: Based on the calcification location change trajectory, the trajectory is connected in time sequence in the image space and associated with the corresponding time information to obtain the calcification migration mapping path.

[0009] As a further aspect of the present invention, the specific steps of S2 are as follows: S201: Based on the calcification migration mapping path, extract the breast ultrasound image contour points of the outer edge of the nodule in the corresponding image frame, sort the contour points according to the spatial position order, analyze the angle change between each point, identify contour segments with bending and acute angle features, and obtain abnormal contour segments. S202: Based on the abnormal contour segments, extract the center position coordinates of each abnormal segment, compare them with the spatial position coordinates of the dense calcification area in the last frame, calculate the spatial overlap ratio between the coordinates, and analyze the distribution relationship of the two types of coordinates in the plane to obtain the relationship between calcification and edge position. S203: Based on the relationship between the calcification and the edge position, determine the contact range boundary between the calcification area and the edge anomaly in the image, convert the spatial range data of the boundary into quantization parameters, and obtain the calcification edge contact distribution structure.

[0010] As a further aspect of the present invention, the specific steps of S3 are as follows: S301: Based on the calcified edge contact distribution structure, divide the nodule region into equal-area gray-scale partitions, analyze the gray-scale value of each gray-scale partition, extract the partitions where the gray-scale value reaches the peak value, and obtain the peak gray-scale partition information. S302: Call the peak grayscale partition information, locate the position of the peak point in the adjacent grayscale partition, calculate the spatial direction vector between the peak points, determine the continuity of the direction, and connect the peak points with continuous directions in sequence to obtain continuous peak path data. S303: Based on the continuous peak path data, describe the extension relationship according to the spatial direction of the path, map the path to the nodule image plane, extract the spatial distribution pattern of the overall extension, and obtain the grayscale structure extension trajectory map.

[0011] As a further aspect of the present invention, the specific steps of S4 are as follows: S401: Based on the grayscale structure extension trajectory map, detect areas with densely distributed extension directions in the nodule image, extract the position coordinates of the areas in the image, and spatially correspond the coordinates with the outer edge of the nodule to obtain the corresponding coordinate set in the extension set; S402: Call the coordinate set corresponding to the extended set, identify the boundary points where the concentrated direction is located in sequence within the outer edge range of the nodule, filter the boundary segments with the same direction, and obtain the information of the concentrated direction boundary segments; S403: Based on the information of the concentrated direction boundary segments, the connected regions between the boundary segments are connected according to spatial continuity, the external contour of the shape is drawn, and the contour is unified to the same spatial reference system to obtain the nodule texture aggregation boundary feature set.

[0012] As a further aspect of the present invention, the specific steps of S5 are as follows: S501: Based on the nodule texture aggregation boundary feature set, the spatial coordinates are matched point by point with the spatial coordinates of the calcified edge contact distribution structure, the feature points that overlap in position are extracted, and grouped according to spatial correlation to obtain the feature overlap point set; S502: Call the feature overlap point set, measure the volume parameters of the corresponding region in the three-dimensional nodule image, and compare them with the external curvature and internal echo distribution values ​​of the corresponding region to extract the difference feature items and obtain the difference feature items. S503: Based on the aforementioned differential feature terms, a feature set that can simultaneously reflect spatial, morphological, and echo characteristics is selected to obtain the nodule classification result.

[0013] A classification system based on thyroid or breast nodules, comprising: The calcification trajectory acquisition module acquires continuous image frames of densely calcified areas in breast or thyroid nodule images, locates the calcified pixel region in each frame, maps it to the spatial location of the image in chronological order, and associates the calcified regions of adjacent frames to obtain the calcification migration mapping path. Based on the calcification migration mapping path, the edge contact recognition module extracts the breast ultrasound image contour points of the outer edge of the nodule in the corresponding image frame, analyzes the adjacent position relationship to identify bending and acute angle segments, and matches the center position of the abnormal segment with the position of the dense calcification area in the last frame to obtain the calcification edge contact distribution structure. The grayscale extension extraction module, based on the calcified edge contact distribution structure, divides the nodule range into equal-area grayscale regions, sorts the grayscale values ​​to locate the peak values, analyzes the extension direction of the peak values ​​in adjacent regions and connects them into a path, describes the spatial extension relationship, and obtains a grayscale structure extension trajectory map. The texture boundary parsing module analyzes the relationship between the concentrated area of ​​the extension direction and the outer edge of the nodule based on the gray structure extension trajectory map, extracts the boundary range of the concentrated direction and draws the connected texture aggregation shape, aligns the boundary features to a unified spatial reference system, and obtains the nodule texture aggregation boundary feature set. The feature classification generation module analyzes the overlapping positions of the calcified edge contact distribution structure based on the nodule texture aggregation boundary feature set, and extracts classification features according to the correspondence between nodule volume, external curvature, and internal echo distribution to obtain the nodule classification result.

[0014] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In this invention, by jointly analyzing the spatial location and temporal changes of calcified regions in continuous image frames, the dynamic migration trajectory is obtained and corresponds to the outer edge features of the nodules. An extension path is constructed by combining the directional distribution of internal grayscale peaks, so that the boundary contact features and texture aggregation patterns are matched in a unified space, forming a multi-dimensional feature set with spatial continuity and directional integrity. This improves the ability to resolve differences in nodule volume, curvature and internal structure, and enhances the stability and discriminability of the classification results. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying 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.

[0016] Figure 1 This is a schematic diagram of the steps of the present invention; Figure 2 This is a detailed schematic diagram of S1 of the present invention; Figure 3 This is a detailed schematic diagram of S2 of the present invention; Figure 4 This is a detailed schematic diagram of S3 of the present invention; Figure 5 This is a detailed schematic diagram of S4 of the present invention; Figure 6 This is a detailed schematic diagram of S5 of the present invention; Figure 7 This is a system module diagram of the present invention. Detailed Implementation

[0017] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0018] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0019] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning.

[0020] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0021] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0022] Please see Figure 1 This invention provides a classification method based on thyroid or breast nodules, comprising the following steps: S1: Obtain a continuous image frame sequence of densely calcified areas in breast or thyroid nodule images, locate the calcified pixel region in each frame, map each frame region to the image spatial location in chronological order, associate the calcified regions of adjacent frames with the continuous change of location over time, and obtain the calcification migration mapping path. S2: Based on the calcification migration mapping path, the breast ultrasound image contour points of the outer edge of the nodule are extracted in the corresponding image frame. The bending and acute angle segments are identified according to the point position order. The center position of the abnormal segment is extracted and spatially correlated with the calcification position in the last frame to obtain the calcification edge contact distribution structure. S3: Based on the calcification edge contact distribution structure, divide the nodule into equal-area gray-scale regions, sort each region by gray-scale, analyze the location of key peaks, determine the peak positions of adjacent regions by direction, and connect them into a continuous extension path. Present the path as an overall extension shape in spatial direction to obtain a gray-scale structure extension trajectory map. S4: Based on the grayscale structure extension trajectory map, analyze the correspondence between the concentrated distribution area of ​​the extension direction and the outer edge of the nodule, extract the boundary range where the concentrated direction is located, draw the connected texture aggregation pattern, align the boundary features of the texture aggregation pattern to a unified spatial reference system, and obtain the nodule texture aggregation boundary feature set. S5: Based on the nodule texture aggregation boundary feature set, the feature position overlap is performed on the calcified edge contact distribution structure. The correspondence between nodule volume, external curvature, and internal echo distribution is analyzed, and the feature set for classification is extracted to obtain the nodule classification result.

[0023] The calcification migration mapping path includes the initial position of the calcification region, the intermediate transition position of the calcification region, and the final position of the calcification region. The calcification edge contact distribution structure includes the position of the edge contact point, the spatial distribution of the contact point, and the density of the contact point. The grayscale structure extension trajectory map includes the distribution of grayscale peak positions, the shape of the peak connection path, and the overall extension direction. The nodule texture aggregation boundary feature set includes the texture aggregation region shape, aggregation boundary line features, and spatial reference system alignment features. The nodule classification results include volume features, external curvature features, and internal echo distribution features.

[0024] Please see Figure 2 The specific steps of S1 are as follows: S101: Obtain a continuous image frame sequence of densely calcified areas in breast or thyroid nodule images, locate the image region of each frame, identify the distribution of calcified pixels, match the identified pixel regions with the spatial coordinates of the frames, and connect the spatial coordinates sequentially in time order to obtain the spatial sequence of calcified pixels. First, the pixel matrix data of each frame in the image source file is retrieved sequentially. The image matrix is ​​read point by point according to the two-dimensional coordinate position. The gray value of each pixel is compared with the average background value of the same frame. When the gray value is higher than the average background value by a set percentage, the pixel is regarded as a calcification candidate point and stored in a temporary set. For example, in a 256×256 ultrasound image, the average background value is 120, and the set percentage is 1.3. Then, pixels with a gray value ≥156 are determined as candidate points. Subsequently, the connectivity of adjacent pixels in the same frame is judged. Candidate points that meet the four-neighbor or eight-neighbor conditions are classified into the same calcification region. For example, the coordinates of the two candidate points in the upper left corner are (34, 45) and (35, 45) respectively. If they are vertically adjacent, they are grouped into the same region. After the regions are merged, the centroid coordinates of each calcified region are extracted and mapped to the two-dimensional spatial position of the current frame. This process is repeated for all frames to obtain the spatial coordinate sequence of calcified regions in each frame. Then, the centroid coordinates of adjacent frames are matched one-to-one according to the frame order. During the matching process, the correspondence with the smallest Euclidean distance between the centroids of adjacent frames is used as the connection basis. For example, if the centroid coordinates of the first frame are (120, 150), the centroid coordinates (122, 151) with the smallest distance are found in the second frame. Then the two are connected as continuous trajectory points. The centroid matching between all frames is completed in sequence and connected into a continuous sequence, finally obtaining the spatial sequence of calcified pixels.

[0025] S102: Based on the spatial sequence of calcified pixels, analyze the spatial coordinate position relationship of calcified regions in adjacent frames, determine the direction of movement of the position in time sequence, connect the continuous spatial change paths in a time continuity manner, and obtain the calcification position change trajectory. First, the centroid coordinates of any two adjacent frames are sequentially retrieved. The centroid coordinates of the previous frame and the next frame are subtracted along the horizontal and vertical axes, respectively. The sign of the difference is determined to identify the horizontal (left-right) and vertical (up-down) movement trends. For example, if the centroid coordinates of the first frame are (120, 150) and the centroid coordinates of the second frame are (123, 148), a horizontal difference of 3 indicates movement to the right, and a vertical difference of -2 indicates movement upwards. This process is repeated for all frames to obtain a sequence of movement directions within each time interval. Then, segments with consecutive directions in the same direction are merged and sequentially linked to form a timeline. Continuous paths, such as three consecutive frames where the horizontal direction is rightward, are directly merged into a single rightward path. The endpoint of this path is then sequentially connected to the starting point of the next path in a different direction to ensure the continuity of spatial position in time. During this process, it is also necessary to judge the situation of abrupt changes in distance. If the centroid Euclidean distance between two adjacent frames exceeds the preset maximum allowable distance, the current path is interrupted and a new path segment is started. Here, the maximum allowable distance can be set according to one-third of the average diameter of the nodules. For example, if the average diameter is 30 pixels, the maximum allowable distance is 10 pixels. Following the above processing flow, all path segments that meet the conditions are sequentially spliced ​​into a whole path in time order, and finally the trajectory of calcification position change is obtained.

[0026] S103: Based on the trajectory of calcification location change, the trajectory is connected in time sequence in the image space and associated with the corresponding time information to obtain the calcification migration mapping path; First, for each spatial coordinate point in the trajectory, connect them sequentially according to their time numbers, linking the coordinate points of the previous time number with the coordinate points of the next time number. During this process, ensure that the direction of the connection is consistent with the chronological order of the time numbers. For example, if the calcification position in the first frame is (130, 145) and in the second frame it is (132, 147), then connect the two points directly with a straight line segment. Assign the time number information of the first and second frames to this connecting segment in the data table. Then, repeat this process for all adjacent frames, gradually forming continuous trajectory lines. Finally, bind each connecting segment to its corresponding time number; for example, the starting time number of a connecting segment might be 5, and the ending time number might be 6. The time information corresponding to this trajectory segment is 5 to 6. After all the connecting segments are bound, a set of spatial path points with time attributes is formed. Then, positional continuity is judged for possible jumps in the trajectory, that is, the distance between the current segment and the end point of the previous segment is compared. When the distance exceeds the allowable range, it is split into two time-continuous paths. The allowable range can be set according to the nodule diameter ratio under the image pixel unit. For example, if the average nodule diameter is 40 pixels, the allowable range can be set to one-quarter of the diameter, that is, 10 pixels. If it exceeds this value, it is broken. Finally, all path segments that meet the continuity condition and have time numbers are connected in order of time number from smallest to largest to obtain the calcification migration mapping path.

[0027] Please see Figure 3 The specific steps of S2 are as follows: S201: Based on the calcification migration mapping path, the breast ultrasound image contour points of the outer edge of the nodule are extracted in the corresponding image frame. The contour points are sorted according to the spatial position order, the angle change between each point is analyzed, and contour segments with bending and acute angle features are identified to obtain abnormal contour segments. First, locate the nodule region contour boundary corresponding to the path in the image frame. Extract continuous contour points from this boundary point by point in the breast ultrasound image. Arrange these contour points sequentially according to their horizontal and vertical coordinates in the image coordinate system. For example, in an image frame, start from the leftmost contour point and connect sequentially to the rightmost contour point to form a point sequence. Then, for any three adjacent points in this point sequence, calculate the angle formed by these three points. Compare all angle values ​​with a predefined bend judgment interval. This interval can be determined based on the smoothness of the nodule edge morphology under normal conditions. For example, set the angle... Points with an angle less than 95 degrees are designated as bend points. The locations of points meeting this condition are recorded separately as candidate bend point locations. The continuity of these candidate bend point locations is then assessed. When multiple candidate points are spatially adjacent and the distance between them does not exceed one-fifth of the average diameter of the nodule, they are merged into a continuous bend segment. Simultaneously, candidate points with an included angle less than 70 degrees are marked as acute angle segments. Acute angle segments and bend segments can overlap, but their location ranges must be recorded separately. Finally, all bend segments and acute angle segments that meet the conditions are mapped in the image frame according to their spatial locations so that they can be directly mapped back to the original image coordinates during subsequent processing to obtain abnormal contour segments.

[0028] S202: Based on the abnormal contour segments, extract the center position coordinates of each abnormal segment, compare them with the spatial position coordinates of the dense calcification area in the last frame, calculate the spatial overlap ratio between the coordinates, and analyze the distribution relationship of the two types of coordinates in the plane to obtain the relationship between calcification and edge position. The formula for calculating the spatial overlap ratio between coordinates specifically represents: ; in, This represents the proportion of spatial overlap between coordinates. This represents the area of ​​overlap between the abnormal contour segment and the calcified region. Unit representing the total area of ​​the abnormal contour segment. This represents the number of coordinate points participating in the comparison. Representing the The horizontal pixel difference between the center of the abnormal segment and the calcification center of the last frame. Representing the The vertical pixel difference between the center of each abnormal segment and the calcification center of the last frame. The unit representing the pixel pitch of an image is millimeters per pixel. The representative is the reference physical distance measured between the tissue effector and the calibration plate. Representing the The directional consistency weight of each corresponding point; The formula's operational logic is as follows: first, calculate the area ratio. , This represents the overlapping area between the abnormal contour segment and the densely calcified area. The total area of ​​the densely calcified region must be represented by the same unit to maintain dimensional consistency. The overall degree of overlap is reflected by the ratio. Then, distance correction calculations are performed on each matched coordinate point, first calculating the pixel coordinate difference between the point in the horizontal and vertical directions. and and multiplied by pixel spacing Convert the distance to physical distance (in mm), then square each point, add them together, and take the square root to obtain the Euclidean distance between the point and the corresponding reference point. Then, use the reference physical distance... The distance to the point is calculated by adding the Euclidean distance to the Euclidean distance as the denominator and using the reference physical distance as the numerator. The two distances are then divided to obtain the distance ratio between the points. This ratio is then weighted with the point's orientation consistency. Multiplication is used as a weighted result, and all Sum the weighted results of each matching point and divide by . The average correction factor is obtained, and finally, the area ratio is multiplied by the average correction factor to obtain the spatial overlap ratio between coordinates. ; Example of parameter acquisition and value setting: The number of intersecting pixels is 820, and the area of ​​each pixel is 0.04 mm². ; If the total number of pixels is 2000, then ; The number of coordinate points involved in the calculation is obtained by matching spatial locations; in this example, we take... ; , The difference between each pair of points in the pixel coordinate system; the first pair of points differs by 5px horizontally and 8px vertically. The pixel spacing during image acquisition, obtained through equipment calibration, is 0.2 mm / px. The reference distance, measured using tissue equivalents, is 2.0 mm. : 1.0, 0.85, 0.75, 0.6; Substitute into the calculation process: Point-by-point calculation: Point 1: Distance = ; Ratio = ; Point 2: Distance = ; Ratio = ; Point 3: Distance = ; Ratio = ; Point 4: Distance = ; Ratio = ; Calculate the average: ; Area ratio: ; result: ; Interpretation of results and numerical significance: The spatial overlap ratio is approximately 0.196, indicating that the degree of spatial overlap between the abnormal contour segment and the dense calcification area is approximately 19.6%. This ratio will be directly used as the basic input data for "relationship between calcification and edge position" at the end of step S202.

[0029] Explanation of the innovative aspects of the formula: The advantage of the formula lies in the introduction of... Reference physical distance and orientation consistency weight The collaborative operation not only takes into account the geometric proximity of the two types of coordinates in space, but also makes the calculation of the spatial overlap ratio reflect both the proximity of positions and the matching of the structural extension trend in space.

[0030] S203: Based on the relationship between calcification and edge location, determine the contact range boundary between the calcification area and edge anomaly in the image, convert the spatial range data of the boundary into quantization parameters, and obtain the calcification edge contact distribution structure. First, the spatial coordinate sets of calcified regions and anomalous edge segments obtained in the previous step are retrieved from the image. These sets are then spatially mapped within the coordinate system of the same image frame. The system detects whether the contours of the calcified regions intersect or are tangent to the contours of the anomalous edges. Next, all outer boundary points of the calcified regions are traversed sequentially to determine whether these points fall within the envelope of the anomalous edge segment. For example, in a nodule image frame, if the horizontal and vertical coordinate differences of the calcified boundary points are both within one pixel, they can be considered to be in contact with the anomalous edge. The coordinates of these contact points are then included in the contact range candidate set. Finally, the candidate sets are grouped according to the connectivity of the contact points. Points that are adjacent to each other and within a distance of no more than 3 pixels are grouped into a contact segment. The coordinates of the four vertices of the bounding rectangle of each contact segment are calculated to define the spatial boundary of the contact segment. The bounding rectangles of all contact segments are then combined to form the overall contact range boundary. Subsequently, the spatial coordinate range of the boundary is converted into actual spatial distance and area parameters by the pixel pitch and the scanning scale. For example, the length and width of the boundary are multiplied by the spatial resolution coefficient of the image to obtain millimeter-level length and width values. The contact area parameters are then calculated by combining the length and width values. These physical quantities such as distance and area are saved as quantization parameters to obtain the calcified edge contact distribution structure.

[0031] Please see Figure 4 The specific steps of S3 are as follows: S301: Based on the calcification edge contact distribution structure, the nodule region is divided into equal-area gray-level partitions, the gray-level value of each gray-level partition is analyzed, and the partitions with gray-level values ​​reaching the peak value are extracted to obtain peak gray-level partition information. First, the outer boundary coordinate set of the nodule region is determined in the two-dimensional pixel matrix of the nodule image, and the region is divided into grids with equal area to obtain grayscale partitions of equal area. For example, a 100×100 pixel nodule region is divided into 25 20×20 pixel partitions. Each partition is individually numbered and corresponds to its spatial coordinates. Then, for each grayscale partition, all pixels in the partition are traversed, the grayscale value of each pixel is extracted, and the average grayscale value of the partition is calculated. At the same time, abnormal grayscale values, such as pixel values ​​below 10 or above 245, are removed during the statistical process to reduce the interference of extreme pixels on the grayscale of the partition. Then, the average grayscale value of all partitions is matched one-to-one with the partition number to form a grayscale value distribution list. The grayscale value of each partition is compared in this list, and the partitions with grayscale values ​​reaching local peaks are selected and their spatial locations are confirmed. For example, when the average grayscale value of a partition is greater than that of the adjacent partitions and the difference exceeds 5 grayscale levels, it can be identified as a peak partition. Finally, the peak grayscale partition information is obtained by assigning the number, spatial coordinate range, and corresponding grayscale value of the peak partition.

[0032] S302: Call the peak grayscale partition information, locate the position of the peak point in the adjacent grayscale partition, calculate the spatial direction vector between the peak points, determine the continuity of the direction, and connect the peak points with continuous directions in sequence to obtain continuous peak path data. The formula for calculating the spatial direction vector between peak points is as follows: ; in, Represents a spatial direction vector. This represents the number of peak points selected during the calculation process. Representing the The difference in x-axis coordinates between each peak point and the previous peak point. Representing the The difference in y-axis coordinates between each peak point and the previous peak point. The unit length calibration value representing the x-axis. The unit length calibration value representing the y-axis; The formula's operational logic: Calculate the coordinate differences of each pair of adjacent peak points on the x-axis and y-axis, denoted as follows: and Next, multiply these two differences to obtain their product. Divide the product by the unit length calibration values ​​of the x-axis and y-axis. and Normalization is performed to unify the scale of data from different coordinate systems. The normalized result is then squared to ensure consistency in the positive and negative directions and improve numerical stability. Finally, all normalized squared values ​​are summed, and the summation is applied to all peak points to obtain the total sum of squares. Finally, the spatial direction vector is obtained by taking the square root. The results show the true spatial distance between the peak points;

[0033] There are two peak points in the image. and ,but: ; The unit calibration values ​​for the x-axis and y-axis are and ; Calibration value ; If there are 10 peak points in an image, then ; Calibration value: , ; Peak point differences: ; Number of peak points: ; Calculate the spatial direction vector: ; Substitute the known values: ; ; ; Interpretation of results and numerical significance: The magnitude of the spatial direction vector between adjacent peak points is approximately 151.66 meters, reflecting the spatial distance relationship between the peak points; Explanation of the innovative aspects of the formula: The advantage of the formula lies in the introduction of normalized calibration values. and The formula can solve the problem of scale differences in differentiated coordinate systems, and by squaring the product of coordinate differences, it can improve the stability and consistency of the calculation. This makes the formula widely applicable in differentiated experimental environments, giving it universality.

[0034] S303: Based on continuous peak path data, the extension relationship is described according to the spatial direction of the path. The path is mapped onto the nodule image plane, and the spatial distribution pattern of the overall extension is extracted to obtain the grayscale structure extension trajectory map. First, the spatial coordinates of each peak point obtained in the previous stage are organized according to the extraction order, and the connection relationship between adjacent peak points is confirmed to ensure path continuity. For example, in a nodule image plane, when the Euclidean distance between adjacent peak points is less than 5 pixels, they can be considered continuous. Then, the direction vectors of each connecting line segment are drawn sequentially according to the spatial direction of the peak path, and these direction vectors are arranged sequentially to form a list of spatial direction descriptions of the path. Each element in this list contains the starting point coordinates, the ending point coordinates, and the direction information. Next, the pixels corresponding to all direction vectors in the path are mapped to the nodule image plane. During the mapping process, the original coordinates need to be normalized so that the position of the path in the image matrix is ​​consistent with the actual imaging position. For example, the original coordinates are scaled to a pixel range of 256×256 according to the image resolution. After the mapping is completed, a linear distribution pattern formed by the path connection is formed in the image plane. Finally, the spatial distribution features of the mapped path on the image plane are extracted based on the overall shape of the path, such as the outline of the pixel area covered by the path, the overall tilt angle range of the direction, and the projection length ratio in the horizontal and vertical directions, to obtain the grayscale structure extension trajectory map.

[0035] Please see Figure 5 The specific steps of S4 are as follows: S401: Based on the grayscale structure extension trajectory map, detect areas with dense distribution of extension direction in nodule images, extract the position coordinates of the areas in the images, and spatially correspond the coordinates with the outer edge of the nodule to obtain the corresponding coordinate set in the extension set; First, the direction vector data of each path in the trajectory map is obtained, and the pixel coordinates corresponding to these direction vectors are marked on the nodule image matrix. Then, a fixed-size window area is gradually slid across the entire image matrix, and the number of direction vectors in each window is counted. Window areas with a number greater than a set reference value are considered as densely distributed areas of extended directions. For example, in a 256×256 pixel image matrix, when the window is 16×16 pixels and contains more than 50 direction vectors, the window position can be determined as a dense area. Next, the coordinates of the upper left and lower right corners of these dense areas in the image matrix are extracted, and these coordinate points are stored according to the region number to form the position coordinate set of the dense areas of extended directions. Then, this coordinate set is spatially matched with the contour point coordinates of the outer edge of the nodule. During the matching process, the shortest pixel distance from each dense area coordinate point to the outer edge contour point is calculated. Coordinate points with a distance of less than 2 pixels are selected as valid corresponding points and classified according to the region number. Finally, the corresponding coordinate set in the extended set is obtained.

[0036] S402: Call the corresponding coordinate set in the extended set, identify the boundary points where the concentrated direction is located in sequence within the outer edge of the nodule, filter the boundary segments with the same direction, and obtain the information of the concentrated direction boundary segments; First, each coordinate point in the coordinate set is located in the pixel matrix of the nodule's outer edge contour. The positions of these coordinate points are retrieved sequentially in a clockwise order according to the outer edge contour points. Then, the direction vector between two adjacent coordinate points is calculated and compared with the reference direction angle. The reference direction angle can be set based on a set of discrete angles with a step size of 30° in the range of 0° to 180°. If the direction angles of two coordinate points fall into the same discrete angle range, they are determined to be consistent. When multiple consecutive coordinate points are consistent in direction, all outer edge pixels between their start and end points are extracted as a boundary segment. For example, in a nodule outer edge contour with a side length of 512 pixels, if the direction angles of 15 consecutive coordinate points are between 90° and 120°, the boundary pixel segment covered by these 15 points can be used as a concentrated direction boundary segment. Then, each boundary segment is numbered and its start and end coordinates and length data are stored to finally obtain the concentrated direction boundary segment information.

[0037] S403: Based on the information of the concentrated direction boundary segments, the connected areas between the boundary segments are connected according to the spatial continuity, the external outline of the shape is drawn, and the outline is unified to the same spatial reference system to obtain the nodule texture aggregation boundary feature set; First, spatial proximity detection is performed on adjacent boundary segments in the pixel coordinate matrix. The detection method is to compare the Euclidean distance between the endpoints of two boundary segments. When the distance value is less than the set spatial continuity reference value (for example, the reference value can be set to 5 pixel intervals), they are considered to have a connection relationship. The endpoints of the two boundary segments are then connected with straight line segments to form a preliminary closed region contour. The above connection operation is repeated among all boundary segments until a set of continuous closed external shape contours is formed. For example, in a 256×256 pixel nodule image, if the endpoint coordinates of boundary segment A are (120, 85) and the starting coordinates of boundary segment B are (123, 88), then the distance between them is about 4.24 pixels, which meets the reference value condition and can be directly connected. After completing all connections, the coordinate system of the obtained contour pixels is uniformly mapped to the global spatial reference system with the upper left corner of the nodule image as the origin, ensuring that the contour coordinates from different sources have uniformity under the same reference datum, and finally the nodule texture aggregation boundary feature set is obtained.

[0038] Please see Figure 6 The specific steps of S5 are as follows: S501: Based on the nodule texture aggregation boundary feature set, the spatial coordinates are matched point by point with the spatial coordinates of the calcified edge contact distribution structure, the feature points that overlap in position are extracted, and grouped according to spatial correlation to obtain the feature overlap point set; First, the two-dimensional coordinates of each feature point are read sequentially from the pixel coordinate dataset, and simultaneously the two-dimensional coordinates of each point in the calcification edge contact distribution data are read. The two are then compared in the same spatial reference system. The comparison process involves calculating the difference between the horizontal and vertical coordinates of a feature point in one of the two data sources and all points in the other dataset. If the difference between the horizontal and vertical coordinates is simultaneously within a set spatial overlap judgment interval (for example, in breast nodule images, this can be set to a deviation of no more than one pixel in each direction), then the point is determined to overlap spatially. In a 256×256 pixel image instance, if the texture... The coordinates of the physical feature point are (105, 89), and the coordinates of the calcification contact point are (106, 89). The difference between their horizontal coordinates is 1, and the difference between their vertical coordinates is 0, which satisfies the overlap condition. This feature point is marked as an overlapping point. After completing the point-by-point matching of all feature points, the resulting overlapping points are aggregated according to their spatial proximity. The proximity is determined by comparing the Euclidean distance between adjacent points. When the distance value does not exceed the set grouping reference value (e.g., 3 pixel spacing), the point is grouped with the adjacent points into the same group. This process is repeated in the coordinate data of the entire image until all overlapping points are grouped, and finally the feature overlapping point set is obtained.

[0039] S502: Call the feature overlap point set, measure the volume parameters of the corresponding region in the 3D nodule image, and compare them with the numerical values ​​of the external curvature and internal echo distribution of the corresponding region to extract the differential feature terms and obtain the differential feature terms. First, the 3D coordinates of each feature overlap point are sequentially located in the voxel data of the 3D nodule image. All voxel units adjacent to this point are extracted, and the boundary range of these voxels is determined as the measurement area corresponding to the feature point. The volume parameter of the measurement area is obtained by counting the number of voxels in the area and multiplying it by the volume of a single voxel. For example, if the side length of a single voxel is 0.5 mm, then the volume of each voxel is 0.125 mm³. If a feature area contains 400 voxels, then the volume of the area is 50 mm³. After the volume measurement is completed, the curvature of the corresponding area is extracted. During curvature measurement, the spatial coordinates of three adjacent boundary points on the surface of the area are selected, and the radius of curvature of the surface formed by the three points at that position is calculated. Then, the reciprocal of the radius of curvature is taken to obtain the curvature value. The curvature values ​​of all boundary points are stored as the measurement area. The data on the curvature distribution of the region is obtained by extracting echo intensity values ​​from ultrasound images within the same feature area. The gray values ​​of each voxel are read and statistically analyzed to obtain the echo distribution data of that area. For example, if the minimum gray value in a certain area is 30, the maximum is 180, and the average is 95, then this set of values ​​is the internal echo distribution value. The measured volume parameters are compared point by point with the curvature distribution data. During the comparison, the distribution ratio of the volume parameters in different curvature intervals is calculated and paired with the echo distribution data for difference calculation. For example, if the volume ratio of a certain area in the curvature interval of 0.02-0.05 is 40%, and the corresponding echo mean is 110, then this ratio and echo value are combined and marked as a difference parameter item. The above measurement, extraction and comparison process is repeated for all feature areas to finally obtain the difference feature item.

[0040] S503: Based on the differential feature terms, a feature set that can simultaneously reflect spatial, morphological and echo characteristics is selected to obtain the nodule classification results; First, the differential feature terms are decomposed according to parameter attributes, and parameters involving spatial characteristics are extracted. These parameters include location coordinates, 3D volume values, and spatial distribution density. For example, location coordinates can be directly read from the (x, y, z) values ​​of a feature point in the 3D image. The 3D volume can be obtained by multiplying the total number of voxels corresponding to the feature by the volume of a single voxel. The spatial distribution density can be calculated by the number of feature points per unit volume. Second, morphological characteristic parameters are extracted. These parameters include the curvature of the outer boundary, the boundary continuity difference, and the degree of surface undulation. The boundary curvature can be calculated from the angle between three consecutive boundary points. The continuity difference can be obtained from the rate of change of the distance between adjacent boundary points. The degree of undulation can be calculated from the height difference of the boundary in the region. Then, the echo characteristic parameters are extracted. These parameters include the mean gray level of the region, the gray level variance, and the skewness of the gray level distribution. The mean gray level can be obtained by averaging the gray levels of all pixels in the region. The mean and mean grayscale variance are averaged by the square of the difference between the grayscale value and the mean. Skewness reflects the skewness of the grayscale value distribution. Then, each of the three types of feature parameters is compared to determine whether the same feature point or feature region simultaneously possesses spatial, morphological, and echo parameters. The comparison must be performed point by point. For example, in a set of three-dimensional data, if the spatial coordinates (35.5, 42.0, 18.0) of a feature point correspond to a volume of 3.25 cubic millimeters, a curvature of 0.015, a grayscale mean of 142, and a variance of 27, then this feature point satisfies the existence conditions of the three attributes. All feature points that meet the conditions are associated and grouped according to their spatial location. When grouping, the Euclidean distance between feature points is measured and compared with a set spatial proximity threshold, for example, the threshold can be 1.0 millimeters. Feature points whose distance does not exceed this value are grouped into the same group. Finally, these features that meet the conditions of spatial, morphological, and echo are grouped to obtain the nodule classification results.

[0041] Please see Figure 7 A classification system based on thyroid or breast nodules, comprising: The calcification trajectory acquisition module acquires continuous image frames of densely calcified areas in breast or thyroid nodule images, locates the calcified pixel region in each frame, maps it to the spatial location of the image in chronological order, and associates the calcified regions of adjacent frames to obtain the calcification migration mapping path. The edge contact recognition module is based on the calcification migration mapping path. It extracts the breast ultrasound image contour points of the outer edge of the nodule in the corresponding image frame, analyzes the adjacent position relationship to identify the bending and acute angle segments, and matches the center position of the abnormal segment with the position of the calcification dense area in the last frame to obtain the calcification edge contact distribution structure. The grayscale extension extraction module is based on the calcification edge contact distribution structure. It divides the nodule range into equal-area grayscale regions, sorts the grayscale values ​​to locate the peak values, analyzes the extension direction of the peak values ​​in adjacent regions and connects them into a path, describes the spatial extension relationship, and obtains the grayscale structure extension trajectory map. The texture boundary parsing module analyzes the relationship between the concentrated area of ​​the extension direction and the outer edge of the nodule based on the gray-scale structure extension trajectory map, extracts the boundary range of the concentrated direction and draws the connected texture aggregation pattern, aligns the boundary features to a unified spatial reference system, and obtains the nodule texture aggregation boundary feature set. The feature classification generation module analyzes the overlapping positions of the calcified edge contact distribution structure based on the nodule texture aggregation boundary feature set, and extracts classification features according to the correspondence between nodule volume, external curvature, and internal echo distribution to obtain the nodule classification result.

[0042] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for classifying thyroid or breast nodules, characterized in that, Includes the following steps: S1: Locate the calcified pixel region in the continuous image frame sequence, map it to the image space coordinates, and associate the homologous calcified segments in adjacent frames according to the continuous change of position over time to obtain the calcification migration mapping path. S2: Based on the calcification migration mapping path, extract the nodule outer edge contour points, identify bending and acute angle segments, and obtain the calcification edge contact distribution structure corresponding to the calcification location space of the last frame. S3: Based on the calcified edge contact distribution structure, divide equal-area gray-scale sub-regions and compare the gray-scale distribution of the sub-regions, locate key peaks, analyze the connectivity of adjacent peaks according to the direction, connect adjacent peaks, present a continuous extension path and overall extension shape, and obtain a gray-scale structure extension trajectory diagram. S4: Based on the grayscale structure extension trajectory map, analyze the correspondence between the concentrated area of ​​the extension direction and the outer edge of the nodule, extract the boundary range of the concentrated direction, draw the texture aggregation pattern, unify the spatial alignment boundary features, and obtain the nodule texture aggregation boundary feature set; S5: Based on the nodule texture aggregation boundary feature set, perform feature position overlap on the calcification edge contact distribution structure, analyze the relationship between nodule volume, external curvature and internal echo distribution, combine discrimination elements, and obtain nodule classification results; The specific steps of S2 are as follows: S201: Based on the calcification migration mapping path, extract the breast ultrasound image contour points of the outer edge of the nodule in the corresponding image frame, sort the contour points according to the spatial position order, analyze the angle change between each point, identify contour segments with bending and acute angle features, and obtain abnormal contour segments. S202: Based on the abnormal contour segments, extract the center position coordinates of each abnormal segment, compare them with the spatial position coordinates of the dense calcification area in the last frame, calculate the spatial overlap ratio between the coordinates, and analyze the distribution relationship of the two types of coordinates in the plane to obtain the relationship between calcification and edge position. S203: Based on the relationship between the calcification and the edge position, determine the contact range boundary between the calcification area and the edge anomaly in the image, convert the spatial range data of the boundary into quantization parameters, and obtain the calcification edge contact distribution structure. The specific steps for S3 are as follows: S301: Based on the calcified edge contact distribution structure, divide the nodule region into equal-area gray-scale partitions, analyze the gray-scale value of each gray-scale partition, extract the partitions where the gray-scale value reaches the peak value, and obtain the peak gray-scale partition information. S302: Call the peak grayscale partition information, locate the position of the peak point in the adjacent grayscale partition, calculate the spatial direction vector between the peak points, determine the continuity of the direction, and connect the peak points with continuous directions in sequence to obtain continuous peak path data. S303: Based on the continuous peak path data, describe the extension relationship according to the spatial direction of the path, map the path to the nodule image plane, extract the spatial distribution pattern of the overall extension, and obtain the grayscale structure extension trajectory map. The homologous calcification segment refers to the process of comparing the centroid position of calcified pixel regions in a continuous image frame sequence according to the time sequence, and then connecting calcified regions with the same position and morphology in adjacent frames in sequence to continue the connection process. The nodule outer edge contour points refer to the set of outer edge coordinate points extracted sequentially along the nodule boundary in the corresponding image frame. Contour segments with bending or acute angle features are identified, and the contact between the calcified area and the nodule edge is analyzed.

2. The classification method based on thyroid or breast nodules according to claim 1, characterized in that, The calcification migration mapping path includes the initial position of the calcification region, the intermediate transition position of the calcification region, and the final position of the calcification region. The calcification edge contact distribution structure includes the edge contact point position, the spatial distribution of contact points, and the density of contact points. The grayscale structure extension trajectory map includes the grayscale peak position distribution, the peak connection path shape, and the overall extension direction. The nodule texture aggregation boundary feature set includes the texture aggregation region shape, aggregation boundary line features, and spatial reference system alignment features. The nodule classification result includes volume features, external curvature features, and internal echo distribution features.

3. The classification method based on thyroid or breast nodules according to claim 1, characterized in that, The calcification location space refers to the coordinate mapping of the initial, intermediate, and final positions of the calcified region in the image plane or three-dimensional space, and the spatial comparison with the coordinates of the outer edge of the nodule to analyze the contact distribution between the calcified region and the edge of the nodule in the differential stage. The texture aggregation morphology refers to finding densely distributed boundary segments in the nodule region based on the grayscale structure extension trajectory map, and sequentially connecting adjacent boundary segments to obtain the outer contour of the texture aggregation region under a unified reference system, reflecting the concentrated distribution state of the echo texture inside the nodule.

4. The classification method based on thyroid or breast nodules according to claim 1, characterized in that, The specific steps of S1 are as follows: S101: Obtain a continuous image frame sequence of densely calcified areas in breast or thyroid nodule images, locate the image region of each frame, identify the distribution of calcified pixels, match the identified pixel regions with the spatial coordinates of the frames, and connect the spatial coordinates sequentially in time order to obtain the spatial sequence of calcified pixels. S102: Based on the calcified pixel spatial sequence, analyze the spatial coordinate position relationship of the calcified region in adjacent frames, determine the direction of movement of the position in the time sequence, connect the continuous spatial change path in the time continuity manner, and obtain the calcified position change trajectory. S103: Based on the calcification location change trajectory, the trajectory is connected in time sequence in the image space and associated with the corresponding time information to obtain the calcification migration mapping path.

5. The classification method based on thyroid or breast nodules according to claim 1, characterized in that, The specific steps of S4 are as follows: S401: Based on the grayscale structure extension trajectory map, detect areas with densely distributed extension directions in the nodule image, extract the position coordinates of the areas in the image, and spatially correspond the coordinates with the outer edge of the nodule to obtain the corresponding coordinate set in the extension set; S402: Call the coordinate set corresponding to the extended set, identify the boundary points where the concentrated direction is located in sequence within the outer edge range of the nodule, filter the boundary segments with the same direction, and obtain the information of the concentrated direction boundary segments; S403: Based on the information of the concentrated direction boundary segments, the connected regions between the boundary segments are connected according to spatial continuity, the external contour of the shape is drawn, and the contour is unified to the same spatial reference system to obtain the nodule texture aggregation boundary feature set.

6. The classification method based on thyroid or breast nodules according to claim 1, characterized in that, The specific steps of S5 are as follows: S501: Based on the nodule texture aggregation boundary feature set, the spatial coordinates are matched point by point with the spatial coordinates of the calcified edge contact distribution structure, the feature points that overlap in position are extracted, and grouped according to spatial correlation to obtain the feature overlap point set; S502: Call the feature overlap point set, measure the volume parameters of the corresponding region in the three-dimensional nodule image, and compare them with the external curvature and internal echo distribution values ​​of the corresponding region to extract the difference feature items and obtain the difference feature items. S503: Based on the aforementioned differential feature terms, a feature set that can simultaneously reflect spatial, morphological, and echo characteristics is selected to obtain the nodule classification result.

7. A classification system based on thyroid or breast nodules, characterized in that, The system is used to implement the classification method based on thyroid or breast nodules as described in any one of claims 1-6, and the system comprises: The calcification trajectory acquisition module acquires continuous image frames of densely calcified areas in breast or thyroid nodule images, locates the calcified pixel region in each frame, maps it to the spatial location of the image in chronological order, and associates the calcified regions of adjacent frames to obtain the calcification migration mapping path. Based on the calcification migration mapping path, the edge contact recognition module extracts the breast ultrasound image contour points of the outer edge of the nodule in the corresponding image frame, analyzes the adjacent position relationship to identify bending and acute angle segments, and matches the center position of the abnormal segment with the position of the dense calcification area in the last frame to obtain the calcification edge contact distribution structure. The grayscale extension extraction module, based on the calcified edge contact distribution structure, divides the nodule range into equal-area grayscale regions, sorts the grayscale values ​​to locate the peak values, analyzes the extension direction of the peak values ​​in adjacent regions and connects them into a path, describes the spatial extension relationship, and obtains a grayscale structure extension trajectory map. The texture boundary parsing module analyzes the relationship between the concentrated area of ​​the extension direction and the outer edge of the nodule based on the gray structure extension trajectory map, extracts the boundary range of the concentrated direction and draws the connected texture aggregation shape, aligns the boundary features to a unified spatial reference system, and obtains the nodule texture aggregation boundary feature set. The feature classification generation module analyzes the overlapping positions of the calcified edge contact distribution structure based on the nodule texture aggregation boundary feature set, and extracts classification features according to the correspondence between nodule volume, external curvature, and internal echo distribution to obtain the nodule classification result.

Images