Intelligent screening system for scoliosis based on three-dimensional reconstruction

Abstract

The present application relates to the technical field of image processing, in particular to a scoliosis intelligent screening system based on three-dimensional reconstruction. The present application obtains the CT image of the current patient through the image acquisition module and extracts the connected domain of the image; further matches the connected domain in the adjacent frame image through the matching module to obtain the connected domain set; further matches the connected domain sets of different axial sections; further extracts the center coordinates of the vertebrae through the auxiliary labeling module for three-dimensional reconstruction; finally analyzes the spatial distribution of the vertebrae, combines the matching coefficients corresponding to the three axial sections, and labels the abnormal vertebrae. The present application extracts the center coordinates of the vertebrae based on three-axis matching and three-dimensional reconstruction, intelligently labels the abnormal vertebrae combined with spatial deviation and matching coefficient, better assists the relevant personnel, effectively overcomes the problem that the scoliosis is difficult to observe due to the deviation of the patient's lying posture, and improves the screening accuracy and robustness.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing, in particular to a scoliosis intelligent screening system based on three-dimensional reconstruction. BACKGROUND

[0002] Scoliosis is a common spinal deformity disease, which often shows that the spine is curved to one side. At present, the screening of scoliosis is usually performed by observing the computed tomography (CT) image, and the computer automatically calculates the curvature degree of the spine according to the obtained image to make a preliminary diagnosis.

[0003] The prior art determines whether scoliosis occurs by automatically calculating the curvature degree of the spine in the obtained CT image and comparing it with a preset threshold. However, in some cases, the patient's posture may affect the spinal image, making the scoliosis not obvious or affected, and thus not easy to observe, which may lead to misjudgment of scoliosis. SUMMARY

[0004] To solve the technical problem that the patient's posture affects the spinal image and makes the scoliosis not easy to observe, the present application aims to provide a scoliosis intelligent screening system based on three-dimensional reconstruction, and the technical solution adopted is as follows:

[0005] Image acquisition module: acquire CT images of three axial sections of the current patient, i.e., horizontal, coronal and sagittal sections, and extract the connected domain of each image;

[0006] Matching module: for each set of axial section images, obtain a matching coefficient and perform connected domain matching according to the distribution consistency between the connected domains of adjacent frames of images, and obtain a connected domain set; match the connected domain sets according to the matching degree of the edge line segments of the connected domains between different axial sections, and combine the similarity of the matching coefficients to match the connected domain sets;

[0007] Auxiliary labeling module: perform three-dimensional reconstruction based on the matched connected domain set, establish a coordinate system and extract the center coordinates of the vertebrae; analyze the spatial distribution of the vertebrae based on the center coordinates of each vertebra, and label the abnormal vertebrae in combination with the matching coefficients corresponding to the three axial sections.

[0008] Further, the method for obtaining the matching coefficient comprises:

[0009] According to the overlapping area, area difference and centroid distance between the connected domains of adjacent frames of images, the matching coefficient between the connected domains is obtained.

[0010] Further, the method for obtaining the connected domain set comprises:

[0011] Sort the matching coefficients of all connected component pairs in adjacent frame images from largest to smallest, and match the connected components according to the sorting order. Each connected component is matched only once in adjacent frame images. Divide the matched connected components of all adjacent frames in the image set into their respective connected component sets.

[0012] Furthermore, the method for matching the set of connected components includes:

[0013] Extract the line segment between any two edge points on each of the connected components, and obtain a connected component tuple with the same length and both ends of the line segment intersecting with two connected components. The connected component tuple belongs to different axial tangents. In any two axial tangents, select one set of connected components for each axial tangent to form a set tuple.

[0014] In the set of binary pairs, the matching degree between two sets of connected components is obtained by combining the proportion of the connected component binary pairs with the similarity of the matching coefficients between the connected components of all the connected component binary pairs.

[0015] The connected component set is matched based on the matching degree.

[0016] Furthermore, the method for matching the set of connected components based on the matching degree includes:

[0017] The connected component set is matched with the connected component set that has the largest matching degree among the connected component sets of the other two axial sections.

[0018] Furthermore, the method for marking abnormal vertebrae includes:

[0019] The vertebrae are sorted from top to bottom according to their relative positions on the human body to construct a sequence; the center coordinates of the first and last vertebrae in the sequence are connected to form the midline of the spine; the distance between the center coordinate of each vertebra and the midline of the spine is obtained as the curvature parameter of each vertebra; in the sequence of vertebrae, the vertebrae corresponding to the maximum value of the curvature parameter are selected one by one as the target vertebra.

[0020] Based on the central tendency of the target vertebra in the sequence, combined with the bending parameters, the matching coefficients corresponding to the three axial sections, and the degree of deviation of the spinal midline from the coordinate axis, the abnormality probability of the target vertebra is obtained; abnormal vertebrae are labeled based on the abnormality probability.

[0021] Furthermore, the method for labeling abnormal vertebrae based on the probability of abnormality includes:

[0022] Vertebrae with an abnormality probability greater than a preset abnormality threshold are identified as abnormal vertebrae and marked.

[0023] Furthermore, the method for establishing the coordinate system includes:

[0024] Establish a coordinate system with the midline of the human body from foot to head as the vertical axis, and any point on the vertical axis as the origin. Use the intersection of the coronal plane and the horizontal plane passing through the origin as the vertical axis, and the intersection of the sagittal plane and the horizontal plane passing through the origin as the horizontal axis.

[0025] Furthermore, the method for obtaining the center coordinates includes:

[0026] Obtain the centroid coordinates of each connected component. For any vertebra, the set of connected components matched by the three axial sections is used. In the set of connected components of the axial section perpendicular to each coordinate axis, the average of the centroid coordinates of the first and last connected components is used as the center coordinates of the corresponding vertebra on each coordinate axis.

[0027] Furthermore, the method for obtaining the connected components includes:

[0028] The grayscale histogram is obtained and the Otsu algorithm is used to obtain the segmentation threshold. The image is then binarized to extract the connected components.

[0029] The present invention has the following beneficial effects:

[0030] This invention first acquires the current patient's CT image and extracts connected components from the image to provide a basis for subsequent analysis. Next, it performs connected component matching in adjacent frames to obtain a set of connected components, resulting in a set of connected components for the same vertebra under a single axial section. Then, based on the degree of matching of the edge segments of the connected component sets across different axial sections, and combined with the similarity of the matching coefficients, the connected component sets are matched to obtain sets of connected components belonging to the same vertebra in each axis. Further, based on the matched connected component sets, three-dimensional reconstruction is performed to extract the center coordinates of the vertebrae, using the distribution of these center coordinates to reflect the geometry of the spine. Finally, the spatial distribution of the vertebrae is analyzed, and abnormal vertebrae are labeled using the matching coefficients corresponding to the three axial sections, improving the identification accuracy of abnormal vertebrae. This invention extracts the center coordinates of vertebrae based on triaxial matching and three-dimensional reconstruction, and intelligently labels abnormal vertebrae by combining spatial deviation and matching coefficients, to better assist relevant personnel and effectively overcome the problem of difficulty in observing scoliosis due to patient lying posture deviations, thus improving the accuracy and robustness of screening. Attached Figure Description

[0031] To more clearly illustrate the technical solutions and advantages 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.

[0032] Figure 1 This is a system block diagram of an intelligent scoliosis screening system based on three-dimensional reconstruction, provided as an embodiment of the present invention.

[0033] Figure 2 An example diagram of a basic human body cross-section and basic axis provided in one embodiment of the present invention;

[0034] Figure 3 An example image of a CT image with three axial sections provided in an embodiment of the present invention;

[0035] Figure 4 An example diagram of connected components in a CT image with three axial sections provided in an embodiment of the present invention;

[0036] Figure 5 This is an example diagram of edge segments matched by connected regions on different cross-sections, provided as an embodiment of the present invention. Detailed Implementation

[0037] 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 three-dimensional reconstruction-based intelligent scoliosis screening system proposed according to 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.

[0038] 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.

[0039] The following description, in conjunction with the accompanying drawings, details a specific scheme for an intelligent scoliosis screening system based on three-dimensional reconstruction provided by the present invention.

[0040] Please see Figure 1 The diagram shows a system block diagram of an intelligent scoliosis screening system based on three-dimensional reconstruction according to an embodiment of the present invention. The system includes: an image acquisition module 101, a matching module 102, and an auxiliary annotation module 103.

[0041] Image acquisition module 101: Acquires CT images of the current patient in three axial sections: horizontal, coronal, and sagittal, and extracts the connected components of the vertebrae in each image.

[0042] Please see Figure 2 It shows an example diagram of a basic human body cross-section and basic axis provided by an embodiment of the present invention. Figure 2The text indicates the upper, lower, anterior (abdomen), posterior (back), and lateral sides of the human body, and shows the horizontal plane (transverse section), sagittal plane (midline), coronal plane (frontal plane), as well as the vertical axis, coronal axis (frontal axis), and sagittal axis.

[0043] Establish a coordinate system with the midline of the human body from foot to head as the vertical axis, and any point on the vertical axis as the origin. Use the intersection of the coronal plane and the horizontal plane passing through the origin as the vertical axis, and the intersection of the sagittal plane and the horizontal plane passing through the origin as the horizontal axis.

[0044] It should be noted that the midline is the midline of the human body, which is the intersection of the sagittal plane (midline) and the coronal plane (frontal plane). The choice of the origin does not affect the relative relationship between the coordinates in the subsequent analysis, so it can be arbitrarily chosen on the vertical axis.

[0045] In one embodiment of the present invention, CT images of three axial sections are acquired, each axial section containing several CT images, and a median filter is used to perform noise reduction processing on the obtained CT images to reduce the influence of noise in the images.

[0046] Furthermore, because bone tissue in the human body has a higher density than other tissues, it absorbs X-rays more effectively. Therefore, the voxels representing bone tissue within a CT scan layer have higher CT values, resulting in higher grayscale values ​​for the corresponding pixels on the CT image after computer conversion. Histogram equalization of the image effectively improves image contrast and enhances the prominence of human bone tissue.

[0047] For the preprocessed CT image, the grayscale histogram is obtained and the segmentation threshold is obtained using the Otsu algorithm. The image is then binarized to extract the connected components.

[0048] Please see Figure 3 This illustration shows an example of a CT image with three axial sections provided in an embodiment of the present invention, corresponding from left to right to horizontal, coronal, and sagittal views, respectively; please refer to... Figure 4 It shows an example diagram of the connected components of a CT image with three axial sections provided in an embodiment of the present invention. Figure 4 correspond Figure 3 The extracted connected components correspond to the horizontal, coronal, and sagittal regions from left to right, respectively.

[0049] It should be noted that the number of CT images for each section is related to the CT scan settings and will not be limited here; the algorithms for CT image acquisition, image preprocessing, and connected component extraction are all existing technologies and will not be elaborated further.

[0050] Matching module 102: For each axial section image set, based on the distribution consistency between connected components of adjacent frame images, matching coefficients are obtained and connected component matching is performed to obtain a connected component set; based on the matching degree of edge segments of connected components between connected component sets of different axial sections, combined with the similarity of matching coefficients, the connected component sets are matched.

[0051] In CT images of the patient's spine taken on different axial planes, although the shape of the same vertebra varies along different axes, its relative position remains consistent across all axial planes. In CT images of the patient's spine taken on several consecutive adjacent planes along the same axis, although the shape and position of the same vertebra may show minor variations, these variations are continuous and follow a predictable pattern. By sequentially matching the connected regions of the same vertebra in each image, the relative position of that vertebra within the patient can be obtained, allowing for three-dimensional reconstruction.

[0052] Therefore, for each image set of axial section, based on the distribution consistency between connected components of adjacent frame images, matching coefficients are obtained and connected component matching is performed to obtain a set of connected components, thus obtaining a set of connected components for the same vertebra under an axial section.

[0053] Preferably, in one embodiment of the present invention, considering that in adjacent frame images, the larger the overlapping area, the smaller the area difference, and the closer the centroid distance, the more likely it is to correspond to the connected domain of the same vertebra and the higher the distribution consistency, the matching coefficient between connected domains is obtained based on the overlapping area, area difference, and centroid distance between adjacent frame images.

[0054] As an example, the numerator is the ratio of the overlapping area of ​​the connected components of adjacent frames to the average area of ​​the two connected components, the absolute value of the difference between the areas of the two connected components is the area difference value, the Euclidean distance between the centroids of the connected components is the centroid distance, the product of the area difference value and the centroid distance is the denominator, and the ratio of the fractions is the matching coefficient of the two connected components.

[0055] The area of ​​connected components is represented by the number of pixels, the data difference is represented by the absolute value of the difference, and the distance between centroids is measured by Euclidean distance. The distribution consistency between connected components is shown from three perspectives: the overlapping area, the area difference, and the centroid distance in adjacent frames.

[0056] Analyze adjacent frame images for each axial section to obtain the matching coefficients between connected components of all adjacent frame images under the same type of section.

[0057] Preferably, in one embodiment of the present invention, the matching coefficients of all connected component pairs in adjacent frame images are sorted from largest to smallest, and the connected components are matched according to the sorting order, with each connected component being matched only once in adjacent frame images; the matched connected components of all adjacent frames in the image set are divided into their respective connected component sets.

[0058] The processing method for consecutive adjacent frames of images in each axial section is consistent, resulting in a set of connected regions belonging to the same vertebra in each axis.

[0059] Considering that for connected regions of the same vertebra on different axial sections, there should be edge segments of equal length. Please refer to... Figure 5 The illustration shows an example diagram of edge segments matched by connected regions on different cross-sections provided by an embodiment of the present invention. Each circular region corresponds to a connected region of the image under a cross-section, and the connecting lines in all circles are the matched common edge segments.

[0060] Meanwhile, considering the similarity of matching coefficients corresponding to connected components with matching edge segments, which reflects the correlation between connected components of different axes, the connected component sets are matched based on the degree of matching of the edges of connected components between connected component sets of different axial sections, combined with the similarity of matching coefficients, to obtain connected component sets of different axial sections corresponding to the same vertebra, providing a basis for subsequent three-dimensional reconstruction.

[0061] Preferably, in one embodiment of the present invention, the line segment between any two edge points on each connected domain is extracted to obtain a connected domain tuple with a common line segment. That is, in the two connected domains, there exists a line segment with the same length, and both ends of the line segment intersect with both connected domains. At the same time, it is ensured that the connected domain tuples belong to different axial tangents to analyze the matching situation of different tangents.

[0062] In any two axial sections, select one connected component set for each axial section to form a set of binary pairs. The analysis process for each set of binary pairs for each of the two axial sections is the same. Here, only one example is described, and it will not be repeated.

[0063] In a set of binary pairs, considering that a higher proportion of connected component binary pairs indicates more connected component pairs with common line segments, and a greater likelihood of corresponding to the same vertebra; and that the more similar the matching coefficients between connected components in a set of binary pairs, the higher the consistency of connected components along different axes, indicating a higher correlation between structures, and a greater likelihood of them being the same vertebra; therefore, based on the proportion of connected component binary pairs and the similarity of the matching coefficients between connected components in all connected component binary pairs, the matching degree between two sets of connected components can be obtained.

[0064] As an example, in a set of binary pairs, the proportion of connected component binary pairs to the total number of connected components is used as the numerator, the sum of the absolute values ​​of the differences in the matching coefficients between the connected components of all connected component binary pairs is used as the denominator, and the ratio of the fractions is used as the matching degree between the corresponding two sets of connected components.

[0065] The proportion of connected component pairs selected by edge segments of the same length that intersect with both connected components demonstrates the matching degree of edge segments between different connected component sets. The data differences are represented by the absolute value of the difference, and the similarity of the matching coefficients is represented by taking the reciprocal and performing a negative correlation mapping.

[0066] Further, the set of connected components is matched based on the matching degree.

[0067] As an example, considering that the degree of matching between connected components of multiple sections belonging to the same vertebra is relatively high, while the degree of matching between connected components of multiple sections belonging to different vertebrae is extremely low due to differences in spatial location and vertebral morphology, the connected component set is matched with the connected component set with the highest matching degree between the connected component sets of the other two axial sections, thus obtaining the connected component set triplet belonging to the same vertebra.

[0068] It should be noted that the length of the edge segment is the Euclidean distance between the two corresponding edge points. In order to avoid the existence of a large number of connected component pairs that meet the requirements due to excessively short common segments, the minimum length of the common segment can be limited, such as a minimum of 3. The matching coefficient of the connected component is the matching coefficient between it and the corresponding connected component of the adjacent previous frame, and the first frame is the matching coefficient between it and the adjacent next frame.

[0069] Auxiliary annotation module 103: Performs 3D reconstruction based on the set of matched connected components, establishes a coordinate system and extracts the center coordinates of the vertebrae; analyzes the spatial distribution of the vertebrae based on the center coordinates of each vertebra, and annotates abnormal vertebrae by combining the matching coefficients corresponding to the three axial sections.

[0070] After matching and obtaining the set of connected components triplets belonging to the same vertebra, three-dimensional reconstruction is performed based on the matched set of connected components. In one embodiment of the present invention, the existing Marching Cubes Algorithm is used to perform three-dimensional reconstruction using the image data of the set of connected components of three different sections to obtain each vertebra of the spine.

[0071] The coordinates of the vertebral centers can form a midline or curve in three-dimensional space, accurately reflecting the geometry of the spine. Therefore, a coordinate system is established and the coordinates of the vertebral centers are extracted. The process of establishing the coordinate system has already been described and will not be repeated here.

[0072] Preferably, in one embodiment of the present invention, the mean value of the centroid coordinates of the first and last slice images in the connected domain set of the axial cross-section perpendicular to each coordinate axis is used as the center coordinate of the corresponding vertebra on each coordinate axis.

[0073] Specifically, the centroid coordinates of each connected component are obtained. For any vertebra, the set of connected components matched by the three axial sections is used. In the set of connected components of the axial section perpendicular to each coordinate axis, the average of the centroid coordinates of the first and last connected components is used as the center coordinates of the corresponding vertebra on each coordinate axis.

[0074] For example, please see Figure 2 Using the sagittal axis as the x-axis, the coronal axis as the y-axis, and the vertical axis as the z-axis, and the vertical axis intersecting the horizontal plane (section), in a set of connected regions corresponding to the horizontal plane, obtain the average of the z-coordinates of the centroids of the highest and lowest connected regions, and use this average as the center coordinates of the corresponding vertebra on the z-axis; similarly, obtain the center coordinates on the other two axes to obtain the center coordinates of the vertebra.

[0075] It should be noted that the method of obtaining the centroid of the connected domain is already existing technology and will not be described in detail here. In other embodiments of the present invention, the implementer may also use the average value of the centroid coordinates of all connected domains in the set of connected domains of the axial section perpendicular to each coordinate axis as the center coordinate of the corresponding vertebra on each coordinate axis.

[0076] Considering that the center coordinates of the vertebrae reflect the geometric shape of the spine in three-dimensional space, and the matching coefficients corresponding to the three axial sections reflect the degree of matching between connected domains and the accuracy of the obtained vertebral center position information, we analyze the spatial distribution of the vertebrae based on the center coordinates of each vertebra, and combine the matching coefficients corresponding to the three axial sections to label abnormal vertebrae, improve the identification of abnormal vertebrae, and avoid the patient's lying posture affecting the spinal image, making scoliosis difficult to observe.

[0077] Preferably, in one embodiment of the present invention, the vertebrae are sorted from top to bottom (from cervical vertebrae to coccyx) relative to the human body position to construct a sequence, and an overall spinal structure model is established to facilitate subsequent coherence judgment and abnormality identification; the center coordinates of the first and last vertebrae of the sequence are connected to form the midline of the spine, which approximately represents the ideal central direction of the normal spine, and provides a benchmark for each vertebra to determine whether it deviates from the overall normal arrangement trend.

[0078] Considering that the greater the distance between the vertebrae and the midline of the spine when the vertebrae are scoliotic, the distance between the center coordinates of each vertebra and the midline of the spine is obtained as the curvature parameter of each vertebra.

[0079] Considering that the maximum value of the bending parameter is the point where the bending is most obvious and corresponds to the key node of the spinal curvature, the vertebrae corresponding to the maximum value of the bending parameter are selected one by one in the vertebral sequence as the target vertebrae to automatically locate potential abnormal locations.

[0080] Considering that the higher the matching coefficient corresponding to the three axial sections, the more accurate the vertebral position and the higher the confidence of the abnormality; considering that the greater the deviation between the midline of the spine and the coordinate axis, the more non-standard the patient's lying posture and the lower the confidence of the abnormality; and considering that scoliosis mostly occurs in the middle segment of the spine, the higher the central tendency of the target vertebra in the sequence, the more likely it is to correspond to an abnormal vertebra.

[0081] Based on this, the probability of an abnormality of the target vertebra is obtained by combining the central tendency of the target vertebra in the sequence with the bending parameters, the matching coefficients corresponding to the three axial sections, and the degree of deviation between the midline of the spine and the coordinate axis; abnormal vertebrae are labeled based on the probability of an abnormality.

[0082] As an example, the absolute value of the difference between the target vertebra's index in the sequence and the median of the index, plus the sum of the sum of the preset positive parameter 0.1 (divided by zero), is taken as the central tendency parameter of the target vertebra, representing the degree of central tendency of the target vertebra in the sequence. The average value of the matching coefficients corresponding to the three axial sections is used as the position confidence, representing the confidence of the spinal vertebra position. The cosine of the acute angle between the spinal midline and the vertical line (the straight line containing the vertical axis) is used as the deviation coefficient. The smaller the angle, the closer the cosine value is to 1, which represents the degree of deviation of the spinal midline from the coordinate axis. It also indicates that the more standard the patient's lying posture, the more accurate the obtained curvature.

[0083] The product of the central tendency parameter, position confidence, deviation coefficient, and curvature parameter is linearly normalized in the corresponding data dimension to obtain the probability of anomaly of the target vertebra. Among them, the spatial distribution of the vertebra is analyzed from multiple perspectives, including the distance between the vertebra and the midline of the spine, the lateral deviation of the midline of the spine, and the central tendency of the target vertebra, and then the probability of anomaly is obtained by fusing the matching coefficient.

[0084] In other embodiments of the present invention, the implementer may also fuse the matching coefficients corresponding to the three axial cross-sections by addition or weighted summation, or fuse the central tendency parameter, position confidence, deviation coefficient and bending parameter, which will not be elaborated further.

[0085] Finally, vertebrae with an abnormal probability greater than the preset abnormal threshold are identified as abnormal vertebrae and marked.

[0086] As an example, the default anomaly threshold is 0.7.

[0087] It should be noted that in this embodiment of the invention, the edge lines of abnormal vertebrae are marked with special marking lines such as dashed lines. The abnormal vertebrae here refer to the abnormal vertebrae in the CT image. This is only to provide auxiliary reference for relevant personnel and does not involve the final judgment of scoliosis.

[0088] In another embodiment of the present invention, the implementer may also treat each vertebra as a target vertebra and analyze all vertebrae; and may also adjust and set other preset abnormal thresholds.

[0089] In summary, addressing the technical problem of scoliosis being difficult to observe due to the patient's lying posture affecting spinal images, this invention proposes an intelligent scoliosis screening system based on three-dimensional reconstruction. This invention acquires the current patient's CT image through an image acquisition module and extracts the connected components of the image; further, a matching module performs connected component matching in adjacent frames to obtain a set of connected components; further, it matches the sets of connected components across different axial sections; further, a three-dimensional reconstruction is performed through an auxiliary annotation module to extract the center coordinates of the vertebrae; finally, the spatial distribution of the vertebrae is analyzed, and abnormal vertebrae are labeled based on the matching coefficients corresponding to the three axial sections. This invention, based on triaxial matching and three-dimensional reconstruction to extract the center coordinates of the vertebrae, and intelligently labels abnormal vertebrae by combining spatial deviation and matching coefficients, better assists relevant personnel, effectively overcomes the problem of scoliosis being difficult to observe due to patient lying posture deviations, and improves the accuracy and robustness of screening.

[0090] 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.

[0091] 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.

Claims

1. A three-dimensional reconstruction based scoliosis intelligent screening system, characterized in that, The system comprises: An image acquisition module: acquiring CT images of three axial sections of the current patient, i.e., horizontal, coronal and sagittal sections, and extracting the connected domains of the vertebrae in each image; A matching module: for each set of images of the axial sections, obtaining a matching coefficient and matching the connected domains according to the distribution consistency between the connected domains of adjacent images, and obtaining a connected domain set; matching the connected domain sets according to the matching degree of the edge segments of the connected domains between different axial sections, and combining the similarity of the matching coefficients to match the connected domain sets; An auxiliary labeling module: performing three-dimensional reconstruction based on the matched connected domain sets, establishing a coordinate system and extracting the center coordinates of the vertebrae; analyzing the spatial distribution of the vertebrae based on the center coordinates of each vertebra, and combining the matching coefficients corresponding to the three axial sections to label abnormal vertebrae; The method for matching the connected domain sets comprises: extracting a line segment between any two edges of each connected domain, obtaining a connected domain pair with the same length of the line segment and intersection points between the two ends of the line segment and two connected domains, and the connected domain pair belonging to different axial sections; in any two axial sections, selecting one connected domain set of each axial section to form a set pair; in the set pair, obtaining the matching degree between the corresponding two connected domain sets according to the proportion of the connected domain pair and the similarity of the matching coefficients between the connected domains of all connected domain pairs; and matching the connected domain sets based on the matching degree; The method for labeling abnormal vertebrae comprises: constructing a sequence by sorting the vertebrae from top to bottom in the relative position of the human body; connecting the center coordinates of the first and last vertebrae in the sequence as the center line of the spine; obtaining the distance between the center coordinates of each vertebra and the center line of the spine as the bending parameter of each vertebra; selecting the vertebra corresponding to the maximum bending parameter in the sequence of vertebrae as the target vertebra; obtaining the abnormality possibility of the target vertebra according to the degree of convergence of the target vertebra in the sequence, the bending parameter, the matching coefficients corresponding to the three axial sections, and the deviation degree of the center line of the spine from the coordinate axis; and labeling the abnormal vertebrae based on the abnormality possibility.

2. The intelligent scoliosis screening system based on three-dimensional reconstruction according to claim 1, wherein, The method for obtaining the matching coefficient comprises: Obtaining the matching coefficient between the connected domains according to the overlapping area, area difference and centroid distance between the connected domains of adjacent images.

3. The intelligent scoliosis screening system based on three-dimensional reconstruction according to claim 1, characterized in that, The method for obtaining the connected domain set comprises: Sorting the matching coefficients of all connected domain pairs of adjacent images from large to small, matching the connected domains according to the sorting order, and matching each connected domain only once in adjacent images; and dividing the matched connected domains of all adjacent images into respective connected domain sets.

4. The intelligent scoliosis screening system based on three-dimensional reconstruction according to claim 1, characterized in that, The method for matching the connected domain sets based on the matching degree comprises: Matching the connected domain sets corresponding to the maximum matching degree between the connected domain sets of the other two axial sections.

5. The intelligent scoliosis screening system based on three-dimensional reconstruction according to claim 1, characterized in that, The method for labeling abnormal vertebrae based on the abnormality possibility comprises: Determining the vertebrae with an abnormality possibility greater than a preset abnormal threshold as abnormal vertebrae and labeling them.

6. The intelligent scoliosis screening system based on three-dimensional reconstruction according to claim 1, wherein, The method for establishing the coordinate system comprises: Taking the median line of the human body from the foot to the head as a vertical axis, taking any point on the vertical axis as an origin, taking the intersection line of the coronal plane and the horizontal plane passing through the origin as a longitudinal axis, and taking the intersection line of the sagittal plane and the horizontal plane passing through the origin as a transverse axis, a coordinate system is established.

7. The intelligent scoliosis screening system based on three-dimensional reconstruction according to claim 6, characterized in that, The method for obtaining the center coordinates comprises: Obtaining the center coordinates of each connected domain, and taking the average of the center coordinates of the first and last connected domains in the connected domain set of the axial section perpendicular to each coordinate axis as the center coordinates of the corresponding vertebra on each coordinate axis. 8.The intelligent scoliosis screening system based on three-dimensional reconstruction of claim 1, wherein, The method for obtaining the connected domain comprises: Obtaining a gray histogram and using an Otsu algorithm to obtain a segmentation threshold, performing binaryzation processing on the image, and extracting the connected domain.

Images