A method and system for extracting imaging features related to glands with endocrine disorders
By extracting features from multi-phase CT scan data and analyzing three-dimensional models, the problem of accuracy in identifying thyroid glands was solved, enabling precise extraction and abnormal assessment of thyroid glands and improving the reliability of diagnosis.
Patent Information
- Application Number
- CN202611256684.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-19
- Publication Date
- 2026-11-17
AI Technical Summary
Existing technologies struggle to accurately segment thyroid glands in different patients and are easily affected by overlapping of adjacent organs, leading to missed or incorrect identification of thyroid glands and impacting diagnostic accuracy.
By acquiring multi-phase CT scan data, combining grayscale distribution features, morphological features, and edge smoothness to calculate probability scores, matching indices and glandular realism, a three-dimensional model is constructed, and the movement trend and abnormal movement index of the glands are analyzed to achieve accurate extraction and evaluation of glandular regions.
It effectively eliminates interference from adjacent organs, improves the accuracy and robustness of thyroid gland extraction, and provides reliable quantitative evidence for clinical diagnosis.
Smart Images

Figure CN123118488A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of image data processing, and specifically to a method and system for extracting image features related to glands with endocrine disorders. Background Technology
[0002] Currently, the identification of imaging features of endocrine abnormalities, such as thyroid cancer, is typically based on medical imaging techniques to identify the abnormal glandular structure. In clinical practice, MRI or multiphase CT scans are often used to identify the specific location, size, and distribution of thyroid adenomas in relation to adjacent structures, enabling healthcare professionals to determine the specific subtype. Existing methods usually extract abnormal glandular features from thyroid images based on fixed models.
[0003] However, in the actual process of identifying abnormal thyroid glands in patients suspected of having thyroid cancer, the following technical problems exist: due to individual differences in the thyroid structure of different patients, including variations in anatomical structures such as physiological protrusions and specific bifurcation morphology, the fixed pre-trained models of existing methods are difficult to accurately segment the thyroid glands of all patients; at the same time, during the scanning process, the overlapping interference of adjacent organs such as lymph nodes may lead to missed or misidentified thyroid glands, affecting the accuracy of diagnosis. Summary of the Invention
[0004] This invention provides a method and system for extracting imaging features related to glands with endocrine abnormalities, in order to solve existing problems.
[0005] The present invention provides a method for extracting imaging features related to glands with endocrine abnormalities, which adopts the following technical solution:
[0006] One embodiment of the present invention provides a method for extracting imaging features related to glands with endocrine abnormalities, the method comprising the following steps:
[0007] Acquire patient scan data from enhanced CT scans, wherein the scan data includes at least two sets, each set including scan data from different scan phases;
[0008] Extract the target region from the scan data, calculate the probability score of each target region based on its grayscale distribution characteristics, morphological characteristics and edge smoothness, and select the suspected gland region from each group of scan data based on the probability score.
[0009] In the same set of scan data, the matching index between suspected gland regions in adjacent images is calculated. Based on the matching index, the suspected gland regions in adjacent images are matched, and the set of continuous images with matching relationships is determined as the target segment.
[0010] Based on the reproducibility of the target segment, the matching residual, and the matching stability between images in different groups of scan data, the gland authenticity of each suspected gland region is calculated, and the suspected gland regions with gland authenticity greater than the preset authenticity threshold are determined as real gland regions.
[0011] Three-dimensional models of the same real glandular region were constructed in different sets of scan data. The consistency between its overall movement direction and local movement direction was analyzed. Combined with the degree of volume change, the abnormal movement index of the real glandular region was calculated.
[0012] Based on gland realism and abnormal movement index, an abnormal image assessment index for the real gland region is calculated, and the real gland region is labeled and output.
[0013] Optionally, target regions are extracted from the scan data. Based on the grayscale distribution characteristics, morphological characteristics, and edge smoothness of each target region, a probability score is calculated for each target region. Suspected gland regions are then selected from each group of scan data based on the probability scores. Specifically, this includes:
[0014] Edge detection is performed on the image in the scanned data, and closed shapes are extracted as target regions;
[0015] For any target region, the gray distribution index of the target region is calculated based on the type of scanning phase, combined with the gray standard deviation, gray mean, overall image gray mean, and contour gradient magnitude.
[0016] For any target region, calculate the edge smoothness index of the target region based on the number of contour points, the maximum length of consecutive points in the same direction in the contour point chain code, and the area of the region.
[0017] The product of the grayscale distribution index and the edge smoothness index is used to determine the probability score that the target region is a gland.
[0018] Target regions with a probability score greater than a preset suspected threshold are identified as suspected glandular regions.
[0019] Optionally, a matching index is calculated between suspected glandular regions in adjacent images, specifically including:
[0020] For any two suspected gland regions in two adjacent images, calculate the probability score difference and spatial overlap area between the two suspected gland regions;
[0021] Normalizing the product of the probability score difference and the spatial overlap area yields the matching index of two suspected glandular regions.
[0022] Optionally, suspected glandular regions in adjacent images are matched based on a matching index, and the set of consecutive images with matching relationships is determined as the target segment, specifically including:
[0023] For two adjacent images, one of them is used as a reference image. For any suspected gland region in the reference image, suspected gland regions that spatially overlap with it are selected in the other image.
[0024] Calculate the matching index between the suspected gland region and each of the screened suspected gland regions, and take the suspected gland region with the largest matching index as its matching region;
[0025] If a matching region exists, then these two suspected gland regions will be marked as belonging to the same target segment;
[0026] Starting from any suspected glandular region in a set of scan data, the matching process is repeated along the forward and reverse directions of the image sequence until no matching region exists.
[0027] The target segment is defined as the set of all successfully matched regions in a continuous image.
[0028] Optionally, based on the reproducibility of the target segment in different groups of scan data, the matching residual, and the matching stability between images, the gland authenticity of each suspected gland region is calculated, specifically including:
[0029] Obtain the matching results between target segments in different groups of scan data. The matching results are determined based on the registration of feature points of the target segments.
[0030] Based on the matching results, the number of times the target segment in the current group of scanned data successfully matches the target segment in the other groups of scanned data is counted as a reproducibility indicator.
[0031] When registering the target segment in the current group of scanned data with the corresponding target segments in other groups of scanned data, obtain the positional error of each feature point after matching.
[0032] The average positional error of all feature points is taken as the average of the matching residuals;
[0033] Obtain the maximum and minimum values of the matching index between adjacent images within the target segment in the current group of scan data, calculate the difference between the two values, and use the matching index range as a characterization of the matching stability between each image.
[0034] Obtain the number of images contained in each target segment corresponding to the suspected gland region in different groups of scan data, and calculate the average number of images;
[0035] The product of the average number of images and the reproducibility index is used as the numerator, and the product of the mean of the matching residuals and the range of the matching index is used as the denominator. The ratio of the numerator to the denominator is calculated to obtain the gland authenticity of the suspected gland region.
[0036] Optionally, obtain the matching results between target segments in different groups of scan data, specifically including:
[0037] For each target segment, obtain the contour points of the suspected glandular region in each of the images it contains;
[0038] For each contour point, obtain the curvature of the contour point, the maximum value of the matching index between the image containing the contour point and its two adjacent images, and the average value of the matching index between all adjacent images within the target segment.
[0039] The first ratio is obtained by dividing the mean of the matching index between all adjacent images in the target segment by the maximum value of the matching index between the image containing the contour point and its two adjacent images.
[0040] Multiply the first ratio by the curvature of the contour point to obtain the probability that the contour point is a feature point;
[0041] Contour points with a probability greater than a preset feature threshold are identified as feature points of the target segment.
[0042] For each target segment in the current group of scan data, the target segment with the most overlapping images with the current target segment is selected as the candidate target segment from the other groups of scan data;
[0043] Extract the image regions that overlap between the current target segment and each candidate target segment, and obtain the feature points on these regions;
[0044] The feature points on the overlapping image region of the current target segment and the candidate target segment are input into the RANSAC algorithm for registration, and the registered matching residual is obtained.
[0045] If the matching residual is less than the preset residual threshold, then the current target segment is determined to be a successful match with the candidate target segment;
[0046] The number of times the current target segment successfully matches the target segments in the other scan data groups is counted to obtain the matching results.
[0047] Optionally, constructing a 3D model of the same real glandular region in different sets of scan data, specifically including:
[0048] For each real gland region, obtain its corresponding target segment in each set of scan data;
[0049] For each target segment in the scan data set, obtain a binary mask of the actual gland region on each of the images it contains;
[0050] Based on the binary masks of each image, an isosurface extraction algorithm is used to reconstruct the three-dimensional model of the real gland region in the set of scan data.
[0051] Optionally, the consistency between the overall and local movement directions is analyzed, and the abnormal movement index of the actual glandular region is calculated in conjunction with the degree of volume change. Specifically, this includes:
[0052] Obtain 3D models of the same real glandular region in different sets of scan data, and determine the coordinates of the center point of each 3D model;
[0053] The direction from the center point of the 3D model in the current group of scan data to the center point of the 3D model in the other groups of scan data is used as the overall movement direction of the real gland region between two scans.
[0054] For each image contained in the target segment in the current group of scan data, obtain the center point coordinates of the real gland region on each image, and use the direction from the center point on the current image to the center point of the real gland region on the corresponding image in the other groups of scan data as the local movement direction of the real gland region on the image.
[0055] Calculate the cosine similarity between the local movement direction on each image and the overall movement direction between scans;
[0056] Obtain the mean and variance of all cosine similarities;
[0057] Obtain the volume of the 3D model of the real gland region in different groups of scan data, and calculate the sum of the pairwise differences in the volume of the 3D model of the real gland region among all scan data.
[0058] Divide the mean of the cosine similarity by the variance of the cosine similarity to obtain the second ratio.
[0059] Dividing the second ratio by the sum of the volume differences yields the abnormal movement index of the actual glandular region.
[0060] Optionally, based on gland realism and abnormal movement index, an abnormal image assessment index for the real gland region is calculated, and the real gland region is labeled and output, specifically including:
[0061] Obtain the gland realism of this real gland region;
[0062] Obtain the abnormal movement index of the actual glandular region;
[0063] Obtain the mean of the abnormal movement index for all real glandular regions;
[0064] Divide the abnormal movement index by the gland authenticity to obtain the first intermediate value;
[0065] Divide the first median value by the mean of all abnormal movement indices of real glandular regions to obtain the abnormal image assessment index of the real glandular region.
[0066] A bounding box is set for the real gland region, and the image features of the real gland region are output in the bounding box. The image features include at least the location, volume and abnormal image evaluation index of the real gland region.
[0067] This invention proposes a system for extracting image features related to glands with endocrine abnormalities, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is executed by the processor, it implements the steps of a method for extracting image features related to glands with endocrine abnormalities.
[0068] The beneficial effects of the technical solution of the present invention are:
[0069] In this embodiment of the invention, by acquiring multi-phase CT scan data of the patient and combining the differences in grayscale distribution and spatial coherence of each region, interference from the identification of adjacent organs such as lymph nodes is effectively eliminated, and the true thyroid gland region is accurately extracted. Simultaneously, based on the movement trend of the gland in the multi-phase scans, the degree of movement rigidity is analyzed, and an abnormal movement index is quantified, thereby comprehensively assessing the morphological and functional abnormalities of the gland. This overcomes the shortcomings of traditional fixed models, which are difficult to adapt to individual structural differences and are easily affected by organ overlap interference, significantly improving the accuracy and robustness of image feature extraction of endocrine abnormal glands, and providing a reliable quantitative basis for clinical diagnosis. Attached Figure Description
[0070] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0071] Figure 1 A flowchart of a method for extracting image features related to glands with endocrine abnormalities, provided in one embodiment of the present invention;
[0072] Figure 2 This is a structural diagram of an image feature extraction system for endocrine abnormal glands provided in an embodiment of the present invention. Detailed Implementation
[0073] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a method for extracting image features related to endocrine abnormalities based on the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.
[0074] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.
[0075] The following describes in detail, with reference to the accompanying drawings, a specific scheme for an image feature extraction method for endocrine abnormal glands provided by the present invention.
[0076] This invention provides a method and system for extracting imaging features related to glands with endocrine abnormalities. Please refer to [link to relevant documentation]. Figure 1 The diagram illustrates a flowchart of an image feature extraction method for endocrine abnormal glands according to an embodiment of the present invention, the method comprising the following steps:
[0077] S101. Acquire the patient's scan data during enhanced CT examination, wherein the scan data includes at least two sets, each set including scan data at different scan phases.
[0078] In this embodiment, the scan data of a single patient suspected of having thyroid cancer are obtained from an enhanced CT scan.
[0079] Multiphase, or multiple scanning phases, refers to performing complete CT scans of the same area of the patient at different time points after contrast agent injection, thereby recording the dynamic changes of contrast agent flow, filling, and dissipation within the tissue. Specifically, multiphase includes:
[0080] Plain scan: The scan before contrast agent injection, used as a baseline reference;
[0081] Arterial phase: Scanning is performed approximately 20-30 seconds after the contrast agent is injected. At this time, the contrast agent is mainly concentrated in the arterial vessels, and tissues with rich blood supply will appear noticeably "brighter".
[0082] Venous phase: Scanning is performed approximately 60-90 seconds after injection, allowing the contrast agent to diffuse into the venous system and interstitial spaces, thus enabling observation of contrast agent uptake in the tissues.
[0083] Delayed phase: Scanning is performed several minutes after injection to observe the rate of contrast agent decay, reflecting the metabolic characteristics of the tissue.
[0084] By combining the scan data collected at different time points, a complete dynamic imaging sequence of the thyroid region under the influence of contrast agent can be obtained, providing a data basis for subsequent analysis of the morphology and position of the gland as it moves with respiration and for excluding interference from adjacent organs.
[0085] For example, this embodiment can collect at least two sets of data including the plain scan phase and the arterial phase, or collect all four sets of data including the plain scan phase, the arterial phase, the venous phase, and the delayed phase, depending on clinical needs and scanning conditions.
[0086] S102. Extract the target region from the scan data. Calculate the probability score of each target region based on its grayscale distribution characteristics, morphological characteristics, and edge smoothness. Based on the probability score, select the suspected glandular regions from each group of scan data.
[0087] In this embodiment, target regions are extracted from the scan data. Based on the grayscale distribution characteristics, morphological characteristics, and edge smoothness of each target region, a probability score is calculated for each target region. Suspected gland regions are then selected from each group of scan data based on the probability scores. Specifically, this includes:
[0088] Edge detection is performed on the image in the scanned data, and closed shapes are extracted as target regions;
[0089] For any target region, the gray distribution index of the target region is calculated based on the type of scanning phase, combined with the gray standard deviation, gray mean, overall image gray mean, and contour gradient magnitude.
[0090] For any target region, calculate the edge smoothness index of the target region based on the number of contour points, the maximum length of consecutive points in the same direction in the contour point chain code, and the area of the region.
[0091] The product of the grayscale distribution index and the edge smoothness index is used to determine the probability score that the target region is a gland.
[0092] Target regions with a probability score greater than a preset suspected threshold are identified as suspected glandular regions.
[0093] For example, the target region is extracted from the scanned data, and a probability score is calculated based on grayscale distribution, shape, and edge features. Specifically, this includes the following steps:
[0094] First, edge detection is performed on the cross-sectional images in each set of scanned data. The Canny operator is used to extract edge information from the images, and adjacent edges are connected by morphological closing operations to form closed shapes. These closed shapes represent regions in the image that may be independent structures, and are used as candidate target regions.
[0095] Thyroid glands are relatively small in cross-sectional images, and their edges are relatively clear on CT images. However, due to their bifurcated structure, the edges appear uneven and rough. By extracting closed shapes through edge detection, all possible structural contours in the image can be initially separated, serving as the basic units for subsequent screening.
[0096] For each target region, its grayscale distribution index is calculated based on the type of scanning phase (the grayscale distribution of thyroid gland tissue on CT images has certain characteristics, namely, its internal grayscale is uneven and there is a significant difference in grayscale compared with the surrounding soft tissues such as fat. Therefore, by analyzing the grayscale distribution within a region, it can be preliminarily determined whether the region is glandular tissue. At the same time, due to the influence of contrast agents in different phases of enhanced CT, the calculation method of grayscale distribution characteristics needs to be different).
[0097] No. The formula for calculating the grayscale distribution index of a target region is:
[0098]
[0099] in, Indicating the first phase of the plain scan image The grayscale distribution index of the target area; Indicates the first The standard deviation of gray level in each target area reflects the degree of gray level unevenness within the area; Indicates the first The average gray level of each target region; This represents the overall grayscale mean of the image; Indicates the first The average gradient magnitude of the region contour of each target region reflects the clarity of the edge. 0.01 is a preset constant to prevent the denominator from being zero.
[0100] The edges of thyroid glands appear rough and uneven on CT images, while the edges of interfering structures such as lymph nodes are often smoother. Quantifying the smoothness of the edges can effectively distinguish glandular and non-glandular structures. Therefore, for the same target region, the edge smoothness index can be calculated using the following formula: ,in, Indicates the first The total number of contour points in the target area reflects the overall length of the contour. Indicates the first The area of each target region is used to normalize the contour length and eliminate the influence of the target region size. Indicates the first The maximum number of consecutive occurrences of the same direction code in the 8-neighborhood chain code of a target region reflects the local smoothness of the contour. The larger this ratio, the smoother the edge of the target region.
[0101] Multiplying the grayscale distribution index by the edge smoothness index yields the first... The probability score of each target region being a gland :
[0102]
[0103] The probability score is calculated by combining the gray-scale features within the target region and the edge morphology features. Only target regions that simultaneously meet the two conditions of "gray-scale distribution conforming to glandular features" and "smooth edges" can obtain a high probability score.
[0104] Then, the probability scores of all target regions in each group of scan data were normalized using Min-Max. Target regions with normalized values greater than a preset suspected threshold were marked as suspected glandular regions. For ease of subsequent description, the selected suspected glandular regions were renumbered. .
[0105] In this embodiment, the preset suspected threshold is set to 0.5. This threshold effectively excludes target areas whose grayscale distribution or edge features clearly do not conform to glandular characteristics while retaining a sufficient number of candidate target areas. It should be noted that this threshold can be adjusted based on actual clinical data and diagnostic needs: if the threshold is set too high, some genuine glandular areas may be missed; if the threshold is set too low, too many interfering target areas will be introduced, increasing the computational burden of subsequent processing. Those skilled in the art can optimize the threshold setting according to specific application scenarios.
[0106] S103. In the same set of scan data, calculate the matching index between suspected gland regions in adjacent images, match the suspected gland regions in adjacent images according to the matching index, and determine the set of continuous images with matching relationship as the target segment.
[0107] In this embodiment, calculating the matching index between suspected glandular regions in adjacent images specifically includes:
[0108] For any two suspected gland regions in two adjacent images, calculate the probability score difference and spatial overlap area between the two suspected gland regions;
[0109] Normalizing the product of the probability score difference and the spatial overlap area yields the matching index of two suspected glandular regions.
[0110] Based on the matching index, suspected glandular regions in adjacent images are matched, and the set of consecutive images with matching relationships is determined as the target segment, specifically including:
[0111] For two adjacent images, one of them is used as a reference image. For any suspected gland region in the reference image, suspected gland regions that spatially overlap with it are selected in the other image.
[0112] Calculate the matching index between the suspected gland region and each of the screened suspected gland regions, and take the suspected gland region with the largest matching index as its matching region;
[0113] If a matching region exists, then these two suspected gland regions will be marked as belonging to the same target segment;
[0114] Starting from any suspected glandular region in a set of scan data, the matching process is repeated along the forward and reverse directions of the image sequence until no matching region exists.
[0115] The target segment is defined as the set of all successfully matched regions in a continuous image.
[0116] In this embodiment, the matching index between suspected glandular regions in adjacent images is first calculated. Then, the suspected glandular regions in adjacent images are matched according to the matching index, and finally the target segment is constructed. The specific steps include:
[0117] For any two suspected glandular regions in two adjacent images, calculate the probability score difference and spatial overlap area between these two suspected glandular regions. Normalize the product of the probability score difference and the spatial overlap area to obtain the... The first suspected glandular area and the first The matching index of a suspected glandular region is denoted as The calculation formula can be:
[0118]
[0119] in, Indicates the first The first suspected glandular area and the first The probability score difference of each suspected glandular region Indicates the first The first suspected glandular area and the first The spatial overlap area of suspected glandular regions. The matching index is used to quantify the probability that two suspected glandular regions in adjacent images belong to the same glandular structure. The smaller the difference in probability scores and the larger the spatial overlap area, the closer the matching index is to 1; conversely, the closer the matching index is to 0. An exponential function is used to ensure that the matching index always remains between 0 and 1.
[0120] The region matching process follows the "maximum value" principle, which means that among candidate regions with spatial overlap, the region with the highest matching index is selected as the matching result. This rule ensures that each suspected gland region has at most one matching region in adjacent images, avoiding confusion in matching relationships.
[0121] Starting from any suspected glandular region in a set of scan data, the above matching process is repeated along the forward and reverse directions of the image sequence until no matching region exists.
[0122] The set of all matching suspected glandular regions in consecutive images is identified as a target segment in the scan data set, denoted as the target segment. .
[0123] For example, if the suspected gland region A in the k-th layer image is successfully matched with the suspected gland region B in the (k+1)-th layer image, and region B is successfully matched with the suspected gland region C in the (k+2)-th layer image, while region C has no matching region in the (k+3)-th layer image, then regions A, B, C and the k, (k+1), and (k+2)-th layers they are in constitute a target segment.
[0124] Through a two-way matching process, all suspected glandular regions of the same glandular structure in consecutive images can be linked together to form a complete three-dimensional structural unit. The target segment represents the complete spatial representation of a suspected glandular structure and is the basic unit for subsequent analysis. Authentic glandular regions usually form longer target segments, while interfering structures such as lymph nodes often form only isolated or shorter target segments.
[0125] S104. Based on the reproducibility of the target segment, the matching residual, and the matching stability between images in different groups of scan data, calculate the gland authenticity of each suspected gland region, and determine the suspected gland region with a gland authenticity greater than the preset true threshold as the true gland region.
[0126] In this embodiment, based on the reproducibility of the target segment in different groups of scan data, the matching residual, and the matching stability between each image, the gland authenticity of each suspected gland region is calculated, specifically including:
[0127] Obtain the matching results between target segments in different groups of scan data. The matching results are determined based on the registration of feature points of the target segments.
[0128] Based on the matching results, the number of times the target segment in the current group of scanned data successfully matches the target segment in the other groups of scanned data is counted as a reproducibility indicator.
[0129] When registering the target segment in the current group of scanned data with the corresponding target segments in other groups of scanned data, obtain the positional error of each feature point after matching.
[0130] The average positional error of all feature points is taken as the average of the matching residuals;
[0131] Obtain the maximum and minimum values of the matching index between adjacent images within the target segment in the current group of scan data, calculate the difference between the two values, and use the matching index range as a characterization of the matching stability between each image.
[0132] Obtain the number of images contained in each target segment corresponding to the suspected gland region in different groups of scan data, and calculate the average number of images;
[0133] The product of the average number of images and the reproducibility index is used as the numerator, and the product of the mean of the matching residuals and the range of the matching index is used as the denominator. The ratio of the numerator to the denominator is calculated to obtain the gland authenticity of the suspected gland region.
[0134] Obtain the matching results between target segments in different groups of scan data, specifically including:
[0135] For each target segment, obtain the contour points of the suspected glandular region in each of the images it contains;
[0136] For each contour point, obtain the curvature of the contour point, the maximum value of the matching index between the image containing the contour point and its two adjacent images, and the average value of the matching index between all adjacent images within the target segment.
[0137] The first ratio is obtained by dividing the mean of the matching index between all adjacent images in the target segment by the maximum value of the matching index between the image containing the contour point and its two adjacent images.
[0138] Multiply the first ratio by the curvature of the contour point to obtain the probability that the contour point is a feature point;
[0139] Contour points with a probability greater than a preset feature threshold are identified as feature points of the target segment.
[0140] For each target segment in the current group of scan data, the target segment with the most overlapping images with the current target segment is selected as the candidate target segment from the other groups of scan data;
[0141] Extract the image regions that overlap between the current target segment and each candidate target segment, and obtain the feature points on these regions;
[0142] The feature points on the overlapping image region of the current target segment and the candidate target segment are input into the RANSAC algorithm for registration, and the registered matching residual is obtained.
[0143] If the matching residual is less than the preset residual threshold, then the current target segment is determined to be a successful match with the candidate target segment;
[0144] The number of times the current target segment successfully matches the target segments in the other scan data groups is counted to obtain the matching results.
[0145] In this embodiment, the matching results between target segments in different groups of scan data are first obtained. Then, the reproducibility index, mean of matching residuals, and range of matching index are calculated based on the matching results. Finally, the glandular authenticity of each suspected glandular region is calculated by combining these parameters. The specific steps include:
[0146] For each target segment, obtain the contour points of the suspected glandular region from the images it contains.
[0147] For the The first target segment From each contour point, obtain the following three parameters:
[0148] No. The first target segment The curvature of a contour point reflects the degree of curvature of the contour at that point. The maximum matching index between the image containing the contour point and its two adjacent images. The mean of the matching index between all adjacent images within the target segment. .
[0149] No. The first target segment The probability that a contour point is a feature point The calculation formula can be:
[0150]
[0151] For all contour points within the target segment Min-Max normalization is performed, and contour points whose normalized values are greater than the preset feature threshold are identified as feature points of the target segment.
[0152] In this embodiment, the preset feature threshold is set to 0.5. This threshold is used to select the most recognizable points on the contour of the target segment as feature points. Feature points need to meet the conditions of large curvature and low matching index of the layer they belong to. The higher the threshold setting, the fewer feature points are selected, but the more representative they are; the lower the threshold setting, the more feature points are selected, but noise may be introduced. Those skilled in the art can optimize and adjust the feature threshold according to image quality and actual needs.
[0153] Feature points are the most identifiable points on the contour of the target segment. They are selected based on the following criteria: given a relatively regular overall structure, the matching index between the layer containing the point and its adjacent layers is small, and the curvature of the point is large. Such points are more easily identified and matched accurately in subsequent cross-group matching.
[0154] For each target segment in the current group of scan data, the target segment with the most overlapping images with the current target segment is selected as the candidate target segment from the remaining groups of scan data.
[0155] Extract the overlapping image regions between the current target segment and each candidate target segment, and obtain the feature points on these overlapping image regions.
[0156] Feature points on the overlapping image regions of the current target segment and the candidate target segment are input into the RANSAC algorithm for registration, yielding the registration matching residual. The matching residual reflects the degree of alignment between the two sets of feature points after spatial transformation; the smaller the residual, the better the match between the two target segments.
[0157] In this embodiment, the preset residual threshold can be set according to image resolution and clinical accuracy requirements, and can be exemplarily set to 2-3 pixels. This threshold is used to determine whether two target segments match successfully: when the matching residual is less than the threshold, the feature points are considered to be well aligned, and the two target segments correspond to the same suspected glandular region; when the matching residual is greater than the threshold, the alignment error is considered too large, and the matching fails. The smaller the threshold setting, the stricter the matching requirements, the fewer the number of successfully matched target segments, but the higher the accuracy; the larger the threshold setting, the higher the matching success rate, but may introduce erroneous matching. Those skilled in the art can optimize the residual threshold setting according to specific application scenarios.
[0158] The purpose of cross-group matching is to determine which target segments correspond to the same suspected glandular region in scans at different respiratory phases or phases. By selecting candidate target segments with the most overlapping images and performing RANSAC registration using feature points, the correspondence between target segments across groups can be effectively established. The number of successful matches will be used to calculate the reproducibility index.
[0159] For each suspected glandular region, obtain the corresponding target segments for that region in different groups of scan data, and perform the following calculations:
[0160] Based on the cross-group matching results, the statistics are as follows: The target segment where the suspected glandular area is located. With the The number of times the target segment was successfully matched in the group scan data is denoted as . This indicator reflects the stability of the suspected glandular region across different sets of scan data.
[0161] In the current group of scan data, the first The target segment where the suspected glandular area is located. When registering with the corresponding target segments in the other scan data sets, the positional error of each feature point after matching is obtained, and the average positional error of all feature points is taken, denoted as . This indicator reflects the degree of morphological similarity between the target segment and the corresponding target segments in other groups; the smaller the mean, the greater the similarity.
[0162] Get the The target segment where the suspected glandular area is located. The maximum and minimum matching indices between adjacent images are calculated, and the difference between them is denoted as . This metric reflects the stability of the matching between different layers of images within the target segment; a smaller range indicates a more stable matching between layers.
[0163] Get the The target segments corresponding to the suspected glandular region in different sets of scan data The average number of images included is denoted as This indicator reflects the spatial extent of the suspected glandular region.
[0164] No. The authenticity of glands in suspected glandular areas It can be calculated as follows:
[0165]
[0166] The design logic for gland realism is as follows: the more likely a region is to be a real gland, the more characteristics it should possess: multiple occurrences in different groups of scan data (high reproducibility), morphological similarity to corresponding target segments in other groups (small matching residuals), stable matching between its own layers (small matching index range), and a wide spatial extension range (large number of images). The first three characteristics are reflected in the denominator, and the latter two characteristics are reflected in the numerator.
[0167] The glandular authenticity of all suspected glandular regions was normalized using Min-Max. Suspected glandular regions with normalized values greater than a preset authenticity threshold were identified as actual glandular regions. For ease of subsequent description, the identified actual glandular regions were renumbered. .
[0168] In this embodiment, the preset true threshold is set to 0.5. This threshold is used to filter out the areas most likely to be actual glandular tissue from the suspected glandular tissue areas. It should be noted that this threshold can be adjusted based on actual clinical data and diagnostic needs: if the threshold is set too high, some actual glandular tissue areas may be missed; if the threshold is set too low, too many interfering areas will be introduced. Those skilled in the art can optimize the threshold setting according to specific application scenarios.
[0169] S105. Construct a three-dimensional model of the same real glandular region in different groups of scan data, analyze the consistency between its overall movement direction and local movement direction, and calculate the abnormal movement index of the real glandular region in combination with the degree of volume change.
[0170] In this embodiment, constructing a three-dimensional model of the same real glandular region in different sets of scan data specifically includes:
[0171] For each real gland region, obtain its corresponding target segment in each set of scan data;
[0172] For each target segment in the scan data set, obtain a binary mask of the actual gland region on each of the images it contains;
[0173] Based on the binary masks of each image, an isosurface extraction algorithm is used to reconstruct the three-dimensional model of the real gland region in the set of scan data.
[0174] The consistency between the overall and local movement directions was analyzed, and the abnormal movement index of the actual glandular region was calculated by combining the degree of volume change. Specifically, this included:
[0175] Obtain 3D models of the same real glandular region in different sets of scan data, and determine the coordinates of the center point of each 3D model;
[0176] The direction from the center point of the 3D model in the current group of scan data to the center point of the 3D model in the other groups of scan data is used as the overall movement direction of the real gland region between two scans.
[0177] For each image contained in the target segment in the current group of scan data, obtain the center point coordinates of the real gland region on each image, and use the direction from the center point on the current image to the center point of the real gland region on the corresponding image in the other groups of scan data as the local movement direction of the real gland region on the image.
[0178] Calculate the cosine similarity between the local movement direction on each image and the overall movement direction between scans;
[0179] Obtain the mean and variance of all cosine similarities;
[0180] Obtain the volume of the 3D model of the real gland region in different groups of scan data, and calculate the sum of the pairwise differences in the volume of the 3D model of the real gland region among all scan data.
[0181] Divide the mean of the cosine similarity by the variance of the cosine similarity to obtain the second ratio.
[0182] Dividing the second ratio by the sum of the volume differences yields the abnormal movement index of the actual glandular region.
[0183] In this embodiment, a three-dimensional model of the same real glandular region is first constructed in different sets of scan data. Then, its movement characteristics and volume changes are analyzed based on the three-dimensional model. Finally, the abnormal movement index of the real glandular region is calculated. The specific steps include:
[0184] For each real gland region, obtain the target segment corresponding to that region in each set of scan data.
[0185] For each target segment in each set of scan data, a binary mask of the actual gland region is obtained on each of the images it contains. A binary mask is a binary image in which the pixel containing the target region is marked as 1, and the remaining pixels are marked as 0.
[0186] Based on the binary masks of each image, an isosurface extraction algorithm is used to reconstruct the three-dimensional model of the real gland region in the set of scan data.
[0187] Binary masks transform region information in a two-dimensional image into pixel-level spatial location markers. Using isosurface extraction algorithms, binary masks from consecutive images can be stacked to reconstruct the three-dimensional morphology of the region. Each real gland region yields a corresponding three-dimensional model in each set of scan data, used for subsequent motion analysis.
[0188] The overall movement direction and the local movement direction are denoted as follows: and The overall movement direction reflects the displacement trend of the actual glandular region as a whole between different respiratory phases or different phases of scanning. For example, from the end of inspiration to the end of expiration, the position of the entire thyroid gland changes within the body, and the direction of displacement of its center point is the overall movement direction. The local movement direction reflects the displacement trend of the actual glandular region in each cross-sectional layer. Because the thyroid gland is a three-dimensional structure, the movement of its different layers during respiration may differ. By calculating the local movement direction on each image, these inter-layer differences can be captured.
[0189] Calculate the similarity between the local movement direction in each image and the overall movement direction across the entire scan to obtain... The cosine similarity value ranges from [-1, 1]: a value of 1 indicates that the two directions are exactly the same, a value of 0 indicates that the two directions are perpendicular, and a value of -1 indicates that the two directions are opposite.
[0190] No. Mean cosine similarity of a real gland region across all images and variance The mean reflects the overall consistency between the local and global movement directions; the variance reflects the fluctuations in the consistency of movement directions between different layers.
[0191] In a normal thyroid gland, the direction of movement of its layers during respiration may differ somewhat from the overall direction of movement, resulting in a non-uniform cosine similarity. However, in a diseased and rigid thyroid gland, the layers may move as a whole, like a rigid body, with the direction of movement of each layer highly consistent with the overall direction of movement (cosine similarity close to 1 and very small variance). This characteristic can be used to determine the degree of abnormality in the gland.
[0192] In volume change analysis, the first The sum of pairwise differences in the volume of the 3D model of a real glandular region between different sets of scan data is denoted as A normal thyroid gland undergoes some stretching during respiration, resulting in a slight change in volume. A diseased, stiff thyroid gland, however, may lose elasticity, leading to smaller volume changes across different respiratory phases. The smaller the sum of these volume differences, the more stable the gland's shape is under different conditions, and the more stiff it may be.
[0193] No. Abnormal movement index of a real glandular region The calculation formula can be:
[0194]
[0195] The design logic of the abnormal movement index is as follows: the higher the degree of abnormality (the more rigid) of a gland, the more it should possess the following characteristics: the movement direction of each layer is highly consistent with the overall movement direction; the fluctuation of the movement direction between layers is small; and the volume change is small under different states. Therefore, The larger, and The smaller the value, the larger the abnormal movement index, indicating that the gland is more rigid and more likely to be diseased tissue.
[0196] In this embodiment, the abnormal movement index is a continuous value and does not involve threshold setting. This index will be used as input for subsequent steps to calculate the final abnormal image assessment index. In actual clinical applications, doctors can combine statistics from a large number of clinical samples to set a reference range or diagnostic threshold for the abnormal movement index to help determine the benign or malignant nature of the gland.
[0197] S106. Based on the gland realism and abnormal movement index, calculate the abnormal image evaluation index of the real gland region and label and output the real gland region.
[0198] In this embodiment, based on gland realism and abnormal movement index, an abnormal image evaluation index for the real gland region is calculated, and the real gland region is labeled and output, specifically including:
[0199] Obtain the gland realism of this real gland region;
[0200] Obtain the abnormal movement index of the actual glandular region;
[0201] Obtain the mean of the abnormal movement index for all real glandular regions;
[0202] Divide the abnormal movement index by the gland authenticity to obtain the first intermediate value;
[0203] Divide the first median value by the mean of all abnormal movement indices of real glandular regions to obtain the abnormal image assessment index of the real glandular region.
[0204] A bounding box is set for the real gland region, and the image features of the real gland region are output in the bounding box. The image features include at least the location, volume and abnormal image evaluation index of the real gland region.
[0205] In calculating the abnormal imaging assessment index, all actual glandular regions in this examination are obtained. The mean of the abnormal movement index For example, if both thyroid lobes are identified as true glandular regions, the average of the two abnormal movement indices is taken; if only one thyroid lobe is identified as a true glandular region, then... This refers to the abnormal movement index of the region itself.
[0206] Abnormal imaging assessment index of real glandular regions The calculation formula can be:
[0207]
[0208] The abnormal image assessment index integrates information from three dimensions:
[0209] A higher abnormal movement index indicates a more rigid gland and a higher degree of abnormality; a lower gland authenticity indicates that the area is less morphologically typical of a gland and may have a greater degree of morphological abnormality; compared with other real gland areas, if its abnormal movement index is higher than the average level, then... This will be amplified further. Therefore, The larger the size, the higher the degree of abnormality of the gland, and the more clinical attention it warrants.
[0210] The abnormal image assessment index of all real glandular regions was normalized using Min-Max normalization to obtain the normalized abnormal image assessment index. Normalization makes the abnormal image assessment index comparable between different patients and different examinations. In this embodiment, the Min-Max normalization method is used.
[0211] It should be noted that in actual clinical applications, a fixed reference range can be set for standardization based on the statistical results of a large number of samples. Those skilled in the art can choose appropriate normalization or standardization methods as needed.
[0212] A bounding box is set for the actual gland region, for example, by drawing a minimum bounding rectangle on the image to enclose the actual gland region. Image features of the actual gland region are output within or near the bounding box, including at least the following information: location information: the spatial coordinates of the actual gland region in the image; volume information: the volume of the 3D model of the actual gland region; anomaly assessment index: a normalized anomaly assessment index that visually reflects the degree of abnormality of the gland.
[0213] The purpose of the labeled output is to present the analysis results to clinicians in a visual form to assist in diagnostic decision-making. The bounding boxes can quickly locate the actual glandular area, while the output image features provide quantitative assessment indicators. The closer the abnormal image assessment index is to 1, the more likely the gland is to be diseased tissue; the closer it is to 0, the more likely it is to be normal.
[0214] In this embodiment, the abnormal image assessment index is a continuous value and no fixed diagnostic threshold is set. In actual clinical applications, doctors can set a reference threshold based on clinical experience and statistical analysis of a large number of samples to assist in judgment.
[0215] In summary, in this embodiment of the invention, through comprehensive analysis of multi-temporal enhanced CT scan data, suspected glandular regions are first screened based on grayscale distribution characteristics, morphological features, and edge smoothness, and target segments are constructed within the same data group based on spatial coherence. Secondly, feature point extraction and RANSAC registration are used to achieve cross-group target segment matching, and the glandular realism is calculated by combining reproducibility, matching residuals, and inter-layer matching stability, effectively eliminating interference from adjacent organs such as lymph nodes. Furthermore, a three-dimensional model is constructed to analyze the consistency of the overall and local movement direction and volume changes of the gland in different temporal phases, quantifying the abnormal movement index. Finally, the abnormal image evaluation index is obtained by combining the glandular realism and the abnormal movement index, and then labeled and output.
[0216] This invention solves the problems of traditional methods being unable to adapt to individual anatomical differences and being susceptible to interference from overlapping organs by using multi-level feature analysis and dynamic behavior quantification. It achieves accurate identification of abnormal endocrine glands and quantitative assessment of the degree of abnormality, providing reliable imaging evidence for clinical diagnosis.
[0217] This invention also proposes a system for extracting imaging features related to glands with endocrine abnormalities. Please refer to [link to relevant documentation]. Figure 2The diagram illustrates a structural diagram of an image feature extraction system for endocrine abnormal glands provided in an embodiment of the present invention. The system includes: a data acquisition module 101, a data processing module 102, and an abnormal output module 103.
[0218] The data acquisition module 101 is used to acquire the patient's scan data in the enhanced CT examination, wherein the scan data includes at least two sets, and each set of data includes scan data at different scan phases;
[0219] The data processing module 102 is used to extract target regions from the scan data, calculate the probability score of each target region based on its grayscale distribution characteristics, morphological characteristics and edge smoothness, and filter out suspected gland regions from each group of scan data based on the probability score.
[0220] In the same set of scan data, the matching index between suspected gland regions in adjacent images is calculated. Based on the matching index, the suspected gland regions in adjacent images are matched, and the set of continuous images with matching relationships is determined as the target segment.
[0221] Based on the reproducibility of the target segment, the matching residual, and the matching stability between images in different groups of scan data, the gland authenticity of each suspected gland region is calculated, and the suspected gland regions with gland authenticity greater than the preset authenticity threshold are determined as real gland regions.
[0222] Three-dimensional models of the same real glandular region were constructed in different sets of scan data. The consistency between its overall movement direction and local movement direction was analyzed. Combined with the degree of volume change, the abnormal movement index of the real glandular region was calculated.
[0223] The abnormal output module 103 is used to calculate the abnormal image evaluation index of the real gland region based on the gland realism and abnormal movement index, and to mark and output the real gland region.
[0224] It should be noted that the system provided in the above embodiments is only an example of the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the computer device can be divided into different functional modules to complete all or part of the functions described above. In addition, the system for extracting image features related to endocrine abnormal glands provided in the above embodiments and the method for extracting image features related to endocrine abnormal glands belong to the same concept. The specific implementation process is detailed in the method embodiments and will not be repeated here.
[0225] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.
[0226] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.
[0227] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A method for extracting image features related to glands with endocrine abnormalities, characterized in that, include: Acquire patient scan data from enhanced CT scans, wherein the scan data includes at least two sets, each set including scan data from different scan phases; Extract the target region from the scan data, calculate the probability score of each target region based on its grayscale distribution characteristics, morphological characteristics and edge smoothness, and select the suspected gland region from each group of scan data based on the probability score. In the same set of scan data, the matching index between suspected gland regions in adjacent images is calculated. Based on the matching index, the suspected gland regions in adjacent images are matched, and the set of continuous images with matching relationships is determined as the target segment. Based on the reproducibility of the target segment, the matching residual, and the matching stability between images in different groups of scan data, the gland authenticity of each suspected gland region is calculated, and the suspected gland regions with gland authenticity greater than the preset authenticity threshold are determined as real gland regions. Three-dimensional models of the same real glandular region were constructed in different sets of scan data. The consistency between its overall movement direction and local movement direction was analyzed. Combined with the degree of volume change, the abnormal movement index of the real glandular region was calculated. Based on gland realism and abnormal movement index, the abnormal image assessment index of the real gland region is calculated, and the real gland region is labeled and output.
2. The method for extracting image features related to endocrine abnormalities according to claim 1, characterized in that, The process involves extracting target regions from the scan data, calculating a probability score for each target region based on its grayscale distribution characteristics, morphological features, and edge smoothness, and then filtering out suspected glandular regions from each set of scan data based on these probability scores. Specifically, this includes: Edge detection is performed on the image in the scanned data, and closed shapes are extracted as target regions; For any target region, the gray distribution index of the target region is calculated based on the type of scanning phase, combined with the gray standard deviation, gray mean, overall image gray mean, and contour gradient magnitude. For any target region, calculate the edge smoothness index of the target region based on the number of contour points, the maximum length of consecutive points in the same direction in the contour point chain code, and the area of the region. The product of the grayscale distribution index and the edge smoothness index is used to determine the probability score that the target region is a gland. Target regions with a probability score greater than a preset suspected threshold are identified as suspected glandular regions.
3. The method for extracting image features related to endocrine abnormalities according to claim 1, characterized in that, The calculation of the matching index between suspected glandular regions in adjacent images specifically includes: For any two suspected gland regions in two adjacent images, calculate the difference in probability scores and the area of spatial overlap between the two suspected gland regions; Normalizing the product of the probability score difference and the spatial overlap area yields the matching index of two suspected glandular regions.
4. The method for extracting image features related to endocrine abnormalities from glands according to claim 1, characterized in that, The step of matching suspected glandular regions in adjacent images based on a matching index, and determining the set of consecutive images with matching relationships as the target segment, specifically includes: For two adjacent images, one of them is used as a reference image. For any suspected gland region in the reference image, suspected gland regions that spatially overlap with it are selected in the other image. Calculate the matching index between the suspected gland region and each of the screened suspected gland regions, and take the suspected gland region with the largest matching index as its matching region; If a matching region exists, then these two suspected gland regions will be marked as belonging to the same target segment; Starting from any suspected glandular region in a set of scan data, the matching process is repeated along the forward and reverse directions of the image sequence until no matching region exists. The target segment is defined as the set of all successfully matched regions in a continuous image.
5. The method for extracting image features related to glands with endocrine abnormalities according to claim 1, characterized in that, The calculation of gland authenticity for each suspected gland region, based on the reproducibility of the target segment in different sets of scan data, matching residuals, and matching stability between images, specifically includes: Obtain the matching results between target segments in different groups of scan data. The matching results are determined based on the registration of feature points of the target segments. Based on the matching results, the number of times the target segment in the current group of scanned data successfully matches the target segment in the other groups of scanned data is counted as a reproducibility indicator. When registering the target segment in the current group of scanned data with the corresponding target segments in other groups of scanned data, obtain the positional error of each feature point after matching. The average positional error of all feature points is taken as the average of the matching residuals; Obtain the maximum and minimum values of the matching index between adjacent images within the target segment in the current group of scan data, calculate the difference between the two values, and use the matching index range as a characterization of the matching stability between each image. Obtain the number of images contained in each target segment corresponding to the suspected gland region in different groups of scan data, and calculate the average number of images; The product of the average number of images and the reproducibility index is used as the numerator, and the product of the mean of the matching residuals and the range of the matching index is used as the denominator. The ratio of the numerator to the denominator is calculated to obtain the gland authenticity of the suspected gland region.
6. The method for extracting image features related to glands with endocrine abnormalities according to claim 5, characterized in that, The acquisition of matching results between target segments in different groups of scan data specifically includes: For each target segment, obtain the contour points of the suspected gland region in each of the images it contains; For each contour point, obtain the curvature of the contour point, the maximum value of the matching index between the image containing the contour point and its two adjacent images, and the average value of the matching index between all adjacent images within the target segment. The first ratio is obtained by dividing the mean of the matching index between all adjacent images in the target segment by the maximum value of the matching index between the image containing the contour point and its two adjacent images. Multiply the first ratio by the curvature of the contour point to obtain the probability that the contour point is a feature point; Contour points with a probability greater than a preset feature threshold are identified as feature points of the target segment. For each target segment in the current group of scan data, the target segment with the most overlapping images with the current target segment is selected as the candidate target segment from the other groups of scan data; Extract the image regions that overlap between the current target segment and each candidate target segment, and obtain the feature points on these regions; The feature points on the overlapping image region of the current target segment and the candidate target segment are input into the RANSAC algorithm for registration, and the registered matching residual is obtained. If the matching residual is less than the preset residual threshold, then the current target segment is determined to be a successful match with the candidate target segment; The number of times the current target segment successfully matches the target segments in the other scan data groups is counted to obtain the matching results.
7. The method for extracting image features related to endocrine abnormalities from glands according to claim 1, characterized in that, The construction of a three-dimensional model of the same real glandular region in different sets of scan data specifically includes: For each real gland region, obtain its corresponding target segment in each set of scan data; For each target segment in the scan data set, obtain a binary mask of the actual gland region on each of the images it contains; Based on the binary masks of each image, an isosurface extraction algorithm is used to reconstruct the three-dimensional model of the real gland region in the set of scan data.
8. The method for extracting image features related to glands with endocrine abnormalities according to claim 1, characterized in that, The analysis of the consistency between the overall and local movement directions, combined with the degree of volume change, calculates the abnormal movement index of the actual glandular region, specifically including: Obtain 3D models of the same real glandular region in different sets of scan data, and determine the coordinates of the center point of each 3D model; The direction from the center point of the 3D model in the current group of scan data to the center point of the 3D model in the other groups of scan data is used as the overall movement direction of the real gland region between two scans. For each image contained in the target segment in the current group of scan data, obtain the center point coordinates of the real gland region on each image, and take the direction from the center point on the current image to the center point of the real gland region on the corresponding image in the other groups of scan data as the local movement direction of the real gland region on the image. Calculate the cosine similarity between the local movement direction on each image and the overall movement direction between scans; Obtain the mean and variance of all cosine similarities; Obtain the volume of the 3D model of the real gland region in different groups of scan data, and calculate the sum of the pairwise differences in the volume of the 3D model of the real gland region among all scan data. Divide the mean of the cosine similarity by the variance of the cosine similarity to obtain the second ratio. Dividing the second ratio by the sum of the volume differences yields the abnormal movement index of the actual glandular region.
9. A method for extracting image features related to glands with endocrine abnormalities according to claim 1, characterized in that, The process involves calculating an abnormal image assessment index for the real gland region based on gland realism and abnormal movement index, and then labeling and outputting the real gland region. Specifically, this includes: Obtain the gland realism of this real gland region; Obtain the abnormal movement index of the actual glandular region; Obtain the mean of the abnormal movement index for all real glandular regions; Divide the abnormal movement index by the gland authenticity to obtain the first intermediate value; Divide the first median value by the mean of all abnormal movement indices of real glandular regions to obtain the abnormal image assessment index of the real glandular region. A bounding box is set for the real gland region, and the image features of the real gland region are output in the bounding box. The image features include at least the location, volume and abnormal image evaluation index of the real gland region.
10. A system for extracting imaging features related to glands with endocrine abnormalities, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the computer program is executed by the processor, it implements the steps of the method for extracting image features related to endocrine abnormal glands as described in any one of claims 1-9.
