Complete imaging method and device for load-bearing position CBCT, electronic equipment and storage medium

Abstract

The invention relates to a load-bearing position CBCT complete imaging method and device, electronic equipment and a storage medium. The method comprises the steps that all scanning areas of CBCT are traversed from bottom to top; acquiring an overlapping region of the adjacent first scanning region and second scanning region; selecting a target main image and a target matching image with the highest matching degree from the main image area and the matching image area, and determining a spatial position corresponding relation; performing image transformation operation on all the images in the second scanning area by adopting a transformation model corresponding to the target matching image; performing fusion operation on the main image region and the transformed matching image region by taking the spatial position corresponding relation as a reference to obtain a fused overlapping region; and splicing the non-overlapping region in the first scanning region, the fused overlapping region and the non-overlapping region in the second scanning region in sequence to obtain a scanning image with continuous axial direction. According to the method and the device, the accuracy of forming a complete image after the load-bearing position CBCT segmented images are spliced is improved.

Description

Technical Field

[0001] This application relates to the field of image technology, and in particular to a complete imaging method, apparatus, electronic device, and storage medium for weight-bearing CBCT. Background Technology

[0002] In the field of orthopedic clinical diagnosis and research, the entire skeletal system, as an interconnected whole, is crucial for disease assessment, particularly the alignment of the spine, hip, knee, and ankle joints. A holistic approach to studying the alignment of the entire skeletal system has become a consensus in the academic community. In clinical practice, whole-spine and whole-lower-limb imaging possess irreplaceable value: whole-spine imaging helps physicians assess the overall spinal structure, accurately identifying problems such as scoliosis, herniated discs, and degenerative spinal diseases; whole-lower-limb imaging is used to examine the health of the hip, knee, and ankle joints and related bones, diagnosing diseases such as arthritis, fractures, and osteoporosis. It also provides a basis for assessing bone growth and development in children and adolescents, enabling timely detection of abnormalities. Therefore, developing three-dimensional imaging technology covering the spine to the hip, knee, and ankle joints in a weight-bearing position, with low radiation and convenient, efficient characteristics, has become a core requirement in orthopedics. Standing imaging, due to its ability to simulate physiological weight-bearing states, plays a decisive role in orthopedic applications.

[0003] Currently, there are two main approaches for full spine and lower limb CT scanning: the traditional supine CT scan and the weight-bearing cone-beam computed tomography (CBCT). Traditional supine CT scans cannot achieve standing scanning, and during routine clinical CT examinations, the examined areas do not bear weight, making it difficult to objectively represent the force distribution and structural stability under standing and weight-bearing physiological conditions. This is especially true in knee, ankle, and spine examinations, where it fails to meet the requirements for simulating normal physiological indicators.

[0004] While weight-bearing CBCT scans attempt to overcome the limitations of traditional CT and are gradually being applied to full spine and lower limb imaging, they face key technical bottlenecks in practical applications. Influenced by factors such as insufficient mechanical precision, imperfect geometric and other correction measures, and differences in structural stability at different heights, weight-bearing CBCT imaging is prone to problems such as stitching seams, image misalignment, and artifacts. These problems not only reduce the subjective visual appeal of the images but, more seriously, lead to errors when using the stitched images for clinical measurements, failing to guarantee the accuracy of the complete image formed by stitching together segmented images. Summary of the Invention

[0005] This application provides a complete imaging method, apparatus, electronic device, and storage medium for heavy-duty CBCT to solve the problem that CBCT cannot guarantee the accuracy of complete imaging after stitching.

[0006] In a first aspect, this application provides a complete imaging method for negative-weight CBCT, the method comprising: All scanned regions obtained from segmented scanning of cone-beam computed tomography are processed sequentially from bottom to top. For the two adjacent scanning regions processed sequentially, a first scanning region located below and a second scanning region located above are determined, and the overlapping region between the first scanning region and the second scanning region is obtained, wherein the lower half of the overlapping region is the main image region and the upper half of the overlapping region is the matching image region; From multiple candidate main images in the main image region and multiple candidate matching images in the matching image region, select the target main image and the target matching image with the highest matching degree and determine the spatial position correspondence. Using the transformation model corresponding to the target matching image, image transformation operations are performed on all images in the second scanning region; Based on the spatial correspondence, a fusion operation is performed on the main image region and the transformed matching image region to obtain the fused overlapping region; The non-overlapping areas in the first scanning area, the fused overlapping area, and the non-overlapping areas in the second scanning area are sequentially stitched together to obtain a continuous scanning image in the axial direction.

[0007] Optionally, selecting the target main image and target matching image with the highest matching degree from multiple candidate main images in the main image region and multiple candidate matching images in the matching image region includes: Multiple candidate main images in the main image region and multiple candidate matching images in the matching image region are cyclically matched to form multiple image matching pairs, wherein each image matching pair is associated with a transformation model; For each image matching pair, interest point detection and matching filtering are performed to determine multiple filtered matching point pairs; Based on the matching point pairs, the optimal transformation model is selected layer by layer from multiple transformation models according to the preset filtering rules; Based on the matching image number associated with the optimal transformation model, the target matching image corresponding to the matching image number is determined, wherein each candidate matching image corresponds to a matching image number, and each matching image number is associated with multiple transformation models. The main image located in the middle position among multiple candidate main images is taken as the target main image.

[0008] Optionally, interest point detection and matching filtering are performed on each of the image matching pairs to determine the multiple filtered matching point pairs, including: For each image matching pair, interest point detection and Lowe ratio test are performed to obtain multiple first matching point pairs after preliminary screening; Based on the distance between the points of interest in each first matching point pair, sort the first matching point pairs in ascending order of distance; Based on the distance distribution of the first matching point pair after sorting, the distance boundary point is determined by combining the distance corresponding to the preset quantile value and the preset proportional coefficient. From the first pair of matching points, select a second pair of matching points whose distance to the point of interest is less than the distance boundary point, and use them as the filtered multiple pairs of matching points.

[0009] Optionally, combining the matching point pairs, the optimal transformation model is selected layer by layer from multiple transformation models according to a preset filtering rule, including: Determine the transformation model associated with each image matching pair and the transformation rules of the transformation model; The second matching point pair in the image matching pair that satisfies the corresponding transformation rule is taken as the inlier. If the number of the inlier is greater than the number threshold, the transformation model associated with the image matching pair is retained and taken as the first transformation model. Determine the transformation matrix corresponding to each first transformation model, and select the second transformation model whose rotation elements are greater than the rotation threshold from multiple transformation matrices; Determine the product of the maximum interior point and the preset scaling parameter in the second transformation model, and filter out the third transformation model with a greater number of interior points than the product value; Based on the matching image sequence number corresponding to each third transformation model, the third transformation model corresponding to the sequence number that appears most frequently is taken as the optimal transformation model.

[0010] Optionally, based on the matching image sequence number corresponding to each third transformation model, the third transformation model corresponding to the sequence number that appears most frequently is selected as the optimal transformation model, including: Determine the matching image number corresponding to each third transformation model; Determine the fourth transformation model associated with the most frequently occurring sequence number; If the number of fourth transformation models exceeds one, the transformation model associated with the candidate matching image closest to the center of the overlapping region is selected as the optimal transformation model.

[0011] Optionally, based on the spatial correspondence, a fusion operation is performed on the main image region and the transformed matching image region to obtain the fused overlapping region, including: Determine the number of overlaps between the main image region and the transformed matching image region; An image index is configured for each of the overlapping images, wherein the image index is used to identify the positional order of the overlapping images from the main image region to the transformed matching image region; A weighted curve is determined based on the number of overlaps and the image index, wherein the weighted curve is monotonically changing and continuously differentiable, and is used to achieve a smooth transition of the overlapping region; Using the weighted curve as the weight, the main image region and the transformed matching image region are weighted and fused to obtain the fused overlapping region.

[0012] Optionally, the formula for calculating the weighted curve is:

[0013] Where w is the weighted curve, N is the number of overlapping images, and j is the image index.

[0014] Secondly, this application provides a complete imaging device for heavy-duty CBCT, the device comprising: The traversal module is used to process all scanned regions obtained by segmented scanning of cone-beam computed tomography in a bottom-up order. The determining module is used to determine a first scanning region located below and a second scanning region located above for two adjacent scanning regions processed sequentially, and to obtain the overlapping area between the first scanning region and the second scanning region, wherein the lower half of the overlapping area is the main image region and the upper half of the overlapping area is the matching image region; The selection module is used to select the target main image and the target matching image with the highest matching degree from multiple candidate main images in the main image region and multiple candidate matching images in the matching image region, and to determine the spatial position correspondence. The image transformation module is used to perform image transformation operations on all images in the second scanning area using the transformation model corresponding to the target matching image; The fusion module is used to perform a fusion operation on the main image region and the transformed matching image region based on the spatial position correspondence to obtain the fused overlapping region. The stitching module is used to sequentially stitch together the non-overlapping areas in the first scanning area, the fused overlapping area, and the non-overlapping areas in the second scanning area to obtain a continuous scanning image in the axial direction.

[0015] Thirdly, this application provides an electronic device, comprising: at least one communication interface; at least one bus connected to the at least one communication interface; at least one processor connected to the at least one bus; and at least one memory connected to the at least one bus.

[0016] Fourthly, this application also provides a computer storage medium storing computer-executable instructions for executing the complete imaging method of weight-bearing CBCT as described in any of the preceding claims of this application.

[0017] The technical solutions provided in this application have the following advantages compared with the prior art: For adjacent first and second scan regions in heavy-weight CBCT segmented scanning, the overlapping area is first divided: the lower half of the overlapping area is designated as the main image region, and the upper half as the matching image region. Then, from multiple candidate main images in the main image region and multiple candidate matching images in the matching image region, the target main image and target matching image with the highest matching degree are selected, and a spatial correspondence between them is established. The transformation model established based on this correspondence can best fit the actual spatial deviation of the second scan region relative to the first scan region. Using the transformation model corresponding to the target matching image, a spatial transformation operation is performed on all images in the second scan region to achieve global spatial alignment between the second and first scan regions, effectively eliminating the misalignment problem between the two regions. Afterwards, based on the spatial correspondence, a weighted fusion process is performed on the main image region and the transformed matching image region to ensure smooth image transition in the overlapping area and avoid obvious stitching marks. After processing adjacent regions in a single group, the images are stitched together in the order of non-overlapping areas of the first scan region, fused overlapping areas, and non-overlapping areas of the second scan region to form a local scan image with continuous axial structure, preventing local breaks from affecting the judgment of the overall skeletal structure. Following this process, the processor processes all segmented scan regions of the CBCT in a bottom-up order, ultimately generating a complete CBCT image with no spatial misalignment, no stitching gaps, and continuous skeletal structure in the load-bearing position, improving the accuracy of forming a complete image after stitching together segmented CBCT images in the load-bearing position. Attached Figure Description

[0018] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0019] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] One or more embodiments are illustrated by way of example with reference numerals in the accompanying drawings. These illustrations do not constitute a limitation on the embodiments. Elements with the same reference numerals in the drawings are denoted as similar elements. Unless otherwise stated, the figures in the drawings are not to be limited by scale.

[0021] Figure 1 A flowchart of a complete imaging method for weight-bearing CBCT provided in this application embodiment; Figure 2 This is a schematic diagram of multi-segment scanning provided for an embodiment of this application; Figure 3 A schematic diagram illustrating a matching and filtering operation using the 0.9 quantile value, provided for an embodiment of this application; Figure 4 This is a schematic diagram of image weighting fusion provided in an embodiment of this application; Figure 5 A schematic diagram of the complete imaging process of weight-bearing CBCT provided in the embodiments of this application; Figure 6 A schematic diagram of the matching process between the candidate main image and the candidate matching image provided in the embodiments of this application; Figure 7 A schematic diagram of a complete imaging device for a weighted CBCT provided in this application embodiment; Figure 8 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0023] The following disclosure provides numerous different embodiments or examples for implementing various structures of this application. To simplify the disclosure, specific examples of components and arrangements are described below. These are merely examples and are not intended to limit the scope of this application. Furthermore, reference numerals and / or letters may be repeated in different examples. Such repetition is for simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or arrangements discussed.

[0024] The following will describe in detail a complete imaging method for negative-position CBCT provided in this application embodiment, taking its application in CBCT as an example. Figure 1 As shown, the specific steps are as follows: Step 101: Process all scanned areas obtained by segmented scanning of cone-beam computed tomography in order from bottom to top; Step 102: For two adjacent scanning regions processed sequentially, determine the first scanning region located below and the second scanning region located above, and obtain the overlapping area between the first scanning region and the second scanning region. The lower half of the overlapping area is the main image region, and the upper half of the overlapping area is the matching image region. Step 103: From multiple candidate main images in the main image region and multiple candidate matching images in the matching image region, select the target main image and target matching image with the highest matching degree and determine the spatial correspondence. Step 104: Using the transformation model corresponding to the target matching image, perform image transformation operation on all images in the second scanning region; Step 105: Based on the spatial correspondence, perform a fusion operation on the main image region and the transformed matching image region to obtain the fused overlapping region; Step 106: Sequentially stitch together the non-overlapping areas in the first scanning area, the fused overlapping area, and the non-overlapping areas in the second scanning area to obtain a continuous scanning image in the axial direction.

[0025] First, some terms used in this application will be explained, including the following:

[0026] Cone-beam computed tomography (CBCT) is an imaging technique that uses a cone-shaped X-ray beam to rotate and scan around the area being examined, combined with data acquired by a detector, and then reconstructs a three-dimensional tomographic image using a computer.

[0027] First scan region: The lower scan region among two adjacent scan regions processed sequentially.

[0028] Second scan region: The scan region located above the first scan region among two adjacent scan regions processed sequentially.

[0029] Overlapping region: The part where the first scanning region and the second scanning region overlap is the core region for achieving image matching and fusion between the two regions. It can be divided into the main image region (lower half) and the matching image region (upper half).

[0030] Candidate main images: Multiple candidate images contained within the main image area (such as images from different angles and frames within the main image area).

[0031] Candidate matching images: Multiple candidate images contained within the matching image area, which correspond one-to-one with the candidate main image to form image matching pairs.

[0032] Target matching image: The image selected from multiple candidate matching images of the matching image region that corresponds to the optimal transformation model.

[0033] Target main image: The image selected from multiple candidate main images in the main image area, located in the middle position.

[0034] Spatial position correspondence: The spatial position mapping relationship between the target main image and the target matching image, such as the coordinate correspondence between a feature point in the target main image and the corresponding feature point in the target matching image.

[0035] Transformation Model: A mathematical model based on image matching pairs, used to realize spatial adjustments such as image rotation, translation, and scaling. Through this model, the entire image of the second scanning area can be adjusted to the same spatial coordinate system as the first scanning area, eliminating the spatial misalignment between the two areas, and also transforming the matching image area.

[0036] Fusion operation: Based on the spatial correspondence, the operation performs weighted processing on the main image region (reference) and the transformed matching image region (already aligned), with the aim of eliminating obvious boundaries (seams) at the splicing of the two regions and achieving a visually smooth transition of the overlapping areas.

[0037] The overlapping region after fusion: The result of the fusion operation, that is, the overlapping part of the main image region and the transformed matching image region after weighted processing, which has no obvious stitching traces.

[0038] Non-overlapping areas: The parts of the first scan area that do not overlap with the second scan area, and the parts of the second scan area that do not overlap with the first scan area, do not require fusion processing and directly participate in the final stitching. They are non-transitional components of the axial continuous scan image.

[0039] Axial continuous scanning image: The image formed by stitching together the non-overlapping area of ​​the first scanning area, the fused overlapping area, and the non-overlapping area of ​​the second scanning area in sequence, is a complete image without breaks or misalignments along the axial direction (such as the vertical direction of the human body). It can fully present the structure and force line status of large-scale orthopedic sites such as the entire spine and the entire lower limb.

[0040] In step 101, in a hospital orthopedic clinical setting, when a doctor performs a CBCT scan of the entire lower limb or spine, the coverage of a single CBCT scan is limited, typically 10-20cm in height. Therefore, multiple local scanning areas need to be obtained through segmented scanning. For example, the entire lower limb can be divided into four areas: ankle to calf, calf to knee, knee to thigh, and thigh to hip. After the scan is completed, the doctor imports the image data of all scanned areas into the image processing system. The image processing system's processor sorts and processes all scanned areas sequentially from bottom to top. Processing from bottom to top ensures that the structural connection between adjacent scanned areas conforms to physiological logic, avoiding errors such as body parts being reversed during subsequent stitching due to disordered order.

[0041] For example, a patient undergoes a full lower limb CBCT scan due to knee pain and abnormal lower limb alignment, resulting in three scan areas: area A (ankle-lower calf, height 15cm), area B (upper calf-knee joint, height 18cm), and area C (knee joint-lower thigh, height 16cm). The image processing system sequentially initiates the processing flow of adjacent areas from bottom to top, starting with area A, area B, and area C.

[0042] In step 102, when the image processing system's processor processes two adjacent scan regions, it first needs to determine the spatial relationship between the two scan regions to ensure subsequent alignment accuracy. The processor determines the lower region among the adjacent regions as the first scan region and the upper region as the second scan region. Then, it extracts the overlapping region where the two regions overlap. Typically, the height of the overlapping region is set to 10%-20% of the height of a single region. This ensures that there are enough points of interest for alignment during subsequent image matching. The processor then divides the overlapping region into two parts: the lower part originates from the first scan region and serves as the main image region, while the upper part originates from the second scan region and serves as the matching image region.

[0043] Figure 2 This diagram illustrates multi-segment scanning. The number of scanning segments, N, is calculated based on the actual imaging range requirements, the range of each scanning segment, and the overlapping area. After planning the number of scanning segments, scanning begins. The machine is moved to the height of the first scanning segment, and the first segment is scanned. After the first segment is completed, the second segment is scanned sequentially, and so on until the Nth segment. It can be seen that in adjacent scanning areas, the lower first scanning area is A2, the upper second scanning area is A3, and the overlapping area between the first and second scanning areas is A6. Similarly, there is also an overlapping area between the first scanning area A4 and the second scanning area A5.

[0044] For example, after a patient's whole lower limb CBCT scan, the adjacent regions are region A (ankle-lower leg) and region B (upper leg-knee joint). The processor determines region A as the first scanning region and region B as the second scanning region, extracts the overlapping region of the two regions at a height of 3cm, and sets the 1.5cm below the overlapping region as the main image region and the 1.5cm above it as the matching image region.

[0045] In step 103, the CBCT scanning area generates multiple consecutive frame images. Therefore, there are multiple candidate main images in the main image area and multiple candidate matching images in the matching image area. The processor of the image processing system performs pairwise cyclic matching of the candidate main images and candidate matching images to obtain multiple image matching pairs. Each image matching pair corresponds to a transformation model. By calculating the interest point matching and filtering the transformation model (i.e., filtering the image matching pairs), the pair with the highest matching degree is finally selected as the target main image and the target matching image, respectively. Then, the processor establishes a spatial correspondence based on the feature point coordinates of these two images, providing a basis for the image transformation of the subsequent second scanning area.

[0046] For example, in a CBCT scan of a patient's entire lower limb, there are 8 candidate main images in the main image region and 8 candidate matching images in the matching image region. The processor, through cyclic matching, identifies and selects the 4th candidate main image and the 3rd candidate matching image as the target main image and target matching image, respectively. A rigid transformation is established whereby the target matching image needs to be translated +2mm along the x-axis, translated -1mm along the y-axis, and rotated 0.4°. Affine transformation and perspective transformation can be added on top of the rigid transformation. The affine transformation is based on the SIFT (Scale-Invariant Feature Transform) algorithm to detect points of interest such as bone edge inflection points and dense trabecular bone areas. Scaling and shearing parameters are added, and a 6-parameter affine matrix is ​​calculated to optimize spatial correspondence. The perspective transformation targets the scan edge region, adds a perspective factor by enhancing feature points, calculates an 8-parameter perspective matrix to correct projection distortion, and finally verifies through reprojection error to ensure a more accurate positional correspondence between the target main image and the target matching image, thus forming a spatial positional correspondence between the target main image and the target matching image.

[0047] In step 104, the target main image and the target matching image form the image matching pair with the highest matching degree. The transformation model associated with this highest matching pair can best fit the actual spatial deviation of the second scanning region relative to the first scanning region, such as region offset or rotation caused by slight patient movement during scanning. Therefore, a unique transformation model associated with this image matching pair can be determined through the target matching image. The processor will apply this transformation model to all images in the second scanning region, because the non-overlapping parts of the second scanning region need to maintain spatial consistency with the overlapping parts to ensure overall stitching continuity. The processor adjusts the coordinates pixel by pixel in each image of the second scanning region, mapping the original coordinates to the target coordinates consistent with the first scanning region, while correcting the rotation angle deviation, to achieve global spatial alignment between the second and first scanning regions, eliminating the misalignment problem between regions. During the transformation process, the matching image region of the second scanning region is also transformed to obtain the transformed matching image region. For example, the second scanning region (region B, upper and middle calf to knee joint) of the patient's whole lower extremity CBCT scan has 20 images. The processor applies a transformation model of x-axis translation +2mm, y-axis translation -1mm, and rotation 0.4° to all images. After the transformation, the anatomical structure of the proximal tibia in the knee joint image in region B is completely aligned with that of the distal tibia in region A. The upper thigh bones in the non-overlapping part also maintain spatial continuity with the overlapping part.

[0048] In step 105, even if the second scanning area has been spatially aligned, the main image area and the transformed matching image area may still have brightness and contrast deviations due to differences in scanning dose. Direct stitching will produce obvious stitching seams, affecting the doctor's judgment of the bone structure. Based on the determined spatial correspondence, the processor performs a fusion operation on the main image area and the transformed matching image area to eliminate stitching marks caused by brightness differences, resulting in a smoothly transitioned fused overlapping area.

[0049] For example, the overlapping area height of the patient's whole lower extremity CBCT scan is 3cm, the average brightness of the main image area is 180, the average brightness of the transformed matching image area is 150, and the brightness of the overlapping area after fusion naturally changes from 180 to 150. The transition between the tibial edge and the trabecular texture is smooth, with no visible splicing traces.

[0050] In step 106, after fusing adjacent regions, the images of each part need to be stitched together to form a complete image of the entire spine or lower limb to meet the needs of doctors to observe the overall skeletal structure. The processor of the image processing system stitches the images of adjacent scanning regions in the order of non-overlapping regions of the first scanning region, overlapping regions after fusion, and non-overlapping regions of the second scanning region. For multiple sets of adjacent regions (such as region B-region C), the processor will repeat the processing of steps 102-105, and finally stitch the images of all scanning regions into a complete scan image of the anatomical structure along the axial direction (vertical direction of the human body). After MPR (Multiplanar Reconstruction) reconstruction or 3D VR display, a complete image of the entire spine or lower limb can be realized.

[0051] For example, a patient's whole lower limb CBCT scan yields three regions. The processor first stitches together the non-overlapping region of region A, the fused overlapping region of region A and region B, and the non-overlapping region of region B. Then, the processing steps 102-105 are repeated for this stitching result and region C (knee joint to lower thigh). Finally, a whole lower limb scan image covering the ankle to the lower thigh is obtained. In the image, the skeletal structures of the ankle, tibia, knee joint, and upper thigh are continuous without misalignment and without any stitching breaks.

[0052] In this application, for adjacent first and second scan regions in a weighted CBCT segmented scan, the overlapping area is first divided, with the lower half designated as the main image region and the upper half as the matching image region. Then, from multiple candidate main images in the main image region and multiple candidate matching images in the matching image region, the target main image and target matching image with the highest matching degree are selected, and a spatial correspondence between them is established. The transformation model adapted based on this correspondence can best approximate the actual spatial deviation of the second scan region relative to the first scan region. Using the transformation model corresponding to the target matching image, a spatial transformation operation is performed on all images in the second scan region, thereby achieving global spatial alignment between the second and first scan regions and effectively eliminating the misalignment problem between the two regions. Afterwards, based on the spatial correspondence, a weighted fusion process is performed on the main image region and the transformed matching image region to ensure smooth image transition in the overlapping area and avoid obvious stitching marks. After processing adjacent regions in a single group, the images are stitched together in the order of non-overlapping areas of the first scan region, fused overlapping areas, and non-overlapping areas of the second scan region to form a local scan image with continuous axial structure, preventing local breaks from affecting the judgment of the overall skeletal structure. Following this process, the processor processes all segmented scan regions of CBCT sequentially from bottom to top, ultimately generating a complete weight-bearing CBCT image without spatial misalignment, stitching gaps, or continuous skeletal structure, thus improving the accuracy of forming a complete image after stitching together segmented weight-bearing CBCT images.

[0053] As an optional implementation, in step 103, selecting the target main image and target matching image with the highest matching degree from multiple candidate main images in the main image region and multiple candidate matching images in the matching image region includes: Step S11: Perform cyclic matching on multiple candidate main images in the main image region and multiple candidate matching images in the matching image region to form multiple image matching pairs, wherein each image matching pair is associated with a transformation model; Step S12: Perform interest point detection and matching filtering for each image matching pair to determine multiple matching point pairs after filtering; Step S13: Combining the matching point pairs, the optimal transformation model is selected layer by layer from multiple transformation models according to the preset filtering rules; Step S14: Determine the target matching image corresponding to the matching image number based on the matching image number associated with the optimal transformation model. Each candidate matching image corresponds to a matching image number, and each matching image number is associated with multiple transformation models. Step S15: Select the main image located in the middle position among multiple candidate main images as the target main image.

[0054] In step S11, the processor performs a pairwise cyclic matching operation on multiple candidate main images (e.g., 5-10 consecutive frame images to reduce scanning motion artifacts) within the main image region and multiple candidate matching images within the matching image region. That is, each candidate main image is combined with all candidate matching images to form multiple candidate main image-candidate matching image image matching pairs. For each image matching pair, the processor establishes a corresponding transformation model (including spatial adjustment parameters such as translation and rotation) based on the preliminary mapping relationship of the anatomical feature points (e.g., bone edges, trabecular texture, joint space markers) between the two, providing a basic candidate set for subsequent selection of the optimal model. For example, when there are 6 candidate main images in the main image region and 8 candidate matching images in the matching image region, the processor can construct 48 image matching pairs and generate a corresponding transformation model for each pair.

[0055] In step S12, the processor first performs interest point detection on the two images in each image matching pair. Interest points in the images specifically refer to key feature points of the skeletal structure in the CBCT image (such as the tibial plateau angle, vertebral body edge, and fibular head apex). These interest points are stable in position during weight-bearing scans and are not affected by soft tissue artifacts. Then, the processor initially matches the interest points in the two images based on their similar descriptions, forming the first matching point pair. Subsequently, the processor performs a matching filtering operation: first, it removes obviously deviated erroneous matching points (such as matching pairs where the Euclidean distance between the two interest points exceeds a preset range) by using a distance threshold; then, it further removes abnormal matching points by using a geometric consistency check (such as checking whether the matching point pair conforms to the spatial continuity of the skeletal structure), finally obtaining multiple filtered matching point pairs, i.e., the second matching point pairs, ensuring that the interest point mapping of each matching pair has anatomical rationality.

[0056] In step S13, the processor uses the second matching point pair as a basis and performs layer-by-layer screening on all transformation models according to preset screening rules: The first layer of screening counts the number of interior points (i.e., matching point pairs that satisfy the model transformation rules) that meet the spatial adjustment parameters of each transformation model, and retains models with more than a preset threshold of interior points to ensure that the model has sufficient reliable matching support; The second layer of screening checks the rationality of the transformation parameters of the retained models, such as whether the rotation angle is within the range of physiological bone activity and whether the translation distance conforms to the scanning area stitching logic, and removes models with abnormal parameters; The third layer of screening sorts the remaining models according to the number of interior points, selects the model with the most interior points and the parameters that best fit the anatomical structure as the optimal transformation model, and if there are models with the same number of interior points, the model with parameters that are closer to the natural spatial state of the bone is selected first.

[0057] In step S14, in this application, each candidate matching image is assigned a unique matching image number, and because the same candidate matching image is combined with different candidate main images, the same matching image number will be associated with multiple transformation models.

[0058] For example, the main image region has 3 candidate main images, denoted as M1, M2, and M3; the matching image region has 2 candidate matching images, which are assigned unique matching image numbers P1 and P2 respectively.

[0059] P1 and M1 are combined (image matching pairs), associated transformation model T1-1; P1 and M2 are combined to form the correlation transformation model T1-2; P1 and M3 are combined to form the correlation transformation model T1-3; P2 and M1 are combined to form the correlation transformation model T2-1; P2 and M2 are combined to form the correlation transformation model T2-2; The combination of P2 and M3 leads to the correlation transformation model T2-3.

[0060] It can be seen that the matching image number P1 is associated with 3 different transformation models, and the matching image number P2 is associated with 3 different transformation models.

[0061] The processor queries the matching image index associated with the optimal transformation model, locates the corresponding candidate matching image using this index, and identifies it as the target matching image. This process avoids the influence of single-model association bias, accurately anchoring the candidate matching image that matches the optimal model through the index, ensuring the reliability of the baseline for subsequent image transformations.

[0062] In step S15, the processor first counts the total number of candidate main images within the main image region, then calculates the middle position. For example, if the total number is odd, the exact middle frame is selected; if the total number is even, the frame after or before the middle two frames is selected, and the candidate main image at that position is determined as the target main image. The reason for selecting the middle position image is that, in the scan frame sequence, the middle frame image is less affected by motion interference at the beginning and end of the scan, and has better image clarity and anatomical feature integrity. Using it as the target main image can reduce the impact of edge frame artifacts on subsequent matching and fusion, ensuring the baseline stability of the overall processing. For example, when there are 7 candidate main images in the main image region, the processor determines the 4th (middle position) candidate main image as the target main image.

[0063] In this application, multiple image matching pairs and corresponding transformation models are constructed through cyclic matching, providing sufficient candidates for optimal model selection and avoiding the selection limitations caused by single pairing. Interest point detection and screening eliminate erroneous matching points, ensuring the anatomy of matching point pairs and laying the foundation for the accuracy of the transformation model. Layer-by-layer screening checks the number of interior points and the rationality of parameters to effectively eliminate inferior transformation models and ensure that the finally selected optimal model can accurately reflect the spatial correspondence between the two regions. The target matching image is accurately located by matching image sequence number to avoid model-image association deviation. The target main image in the middle position is selected to reduce edge frame artifact interference and ensure the stability of the matching benchmark.

[0064] As an optional implementation, in step S12, interest point detection and matching filtering are performed on each image matching pair, and the multiple matching point pairs after filtering are determined as follows: Step S121: Perform interest point detection and Lowe ratio test on each image matching pair to obtain multiple first matching point pairs after preliminary screening; Step S122: Sort the first matching point pairs in ascending order of distance based on the distance between the points of interest in each first matching point pair; Step S123: Based on the distance distribution of the first matching point pair after sorting, determine the distance boundary point by combining the distance corresponding to the preset quantile value and the preset proportional coefficient; Step S124: From the first pair of matching points, select the second pair of matching points whose distance to the point of interest is less than the distance boundary point, and use them as the filtered multiple pairs of matching points.

[0065] In step S121, the processor first detects interest points (OPPs) in both images for each candidate main image-candidate matching image image matching pair and generates a feature description for each OPP. Then, the processor performs a Lowe ratio test using OPP matching pair filtering method 1 (C4): for any OPP in each candidate main image (denoted as main point X), it finds the two OPPs (denoted as matching point A and matching point B) in the candidate matching image that have the highest similarity to its feature description. It calculates the distance (d1) between main point X and matching point A, and the distance (d2) between main point X and matching point B, and calculates the distance ratio (d1 / d2). If the distance ratio exceeds the empirical threshold of 0.7, the processor determines that matching point B is an interference point, and the matching pair (X+B) is too heavily affected by interference, resulting in a low probability of accurate matching; therefore, it needs to be discarded. If the distance ratio is <0.7, the processor determines that matching point A has stronger consistency in anatomical features with main point X and retains the matching pair (X+A). Through this process, the processor performs interest point detection and Lowe ratio test on each image matching pair, obtaining multiple first matching point pairs after preliminary screening, effectively eliminating erroneous matches caused by image noise (such as metal artifacts in CBCT scans).

[0066] In step S122, the processor calculates the Euclidean distance in three-dimensional space for each pair of interest points (e.g., principal point X and matching point A) corresponding to the first matching point pair of each image matching pair. This distance reflects the spatial fit between the two interest points; the smaller the distance, the higher the accuracy of the anatomical structure correspondence. Subsequently, the processor sorts all the first matching point pairs of the same image matching pair in ascending order of distance, based on the distance between interest points, forming an ordered sequence of matching point pairs. This lays the data foundation for subsequent selection of the optimal matching point pairs based on distance distribution, facilitating the rapid location of reasonable intervals and discrete values ​​in the distance distribution.

[0067] In step S123, the processor analyzes the distance distribution characteristics based on the sorted first matching point pair distance sequence using interest point matching pair screening method 2 (C5): First, it calculates the distance corresponding to the preset quantile value (e.g., 0.9 quantile) of the sequence using statistical tools, denoted as L. Selecting the preset quantile value can effectively avoid extremely discrete values ​​in the distance sequence (e.g., abnormal matching point pair distances caused by blurred skeleton edges). Practice has shown that compared to directly taking the maximum value of the distance sequence, the preset quantile value can more accurately reflect the distance range of most reasonable matching point pairs, reducing the interference of extreme values ​​on screening. Subsequently, the processor combines a preset proportional coefficient (e.g., coefficient is 0.75) to calculate the distance boundary point as the preset proportional coefficient * L. This boundary point retains matching point pairs with small distances and high fit, and provides a clear standard for subsequently eliminating matching point pairs with slight deviations.

[0068] Figure 3 This diagram illustrates the matching and filtering operation using the 0.9 quantile. It can be seen that points of interest before the 0.9 quantile have a high number of matching point pairs.

[0069] In step S124, the processor traverses the sorted sequence of first matching point pairs, comparing the distance of the point of interest (POI) of each matching point pair with a determined distance boundary (0.75*L). If the POI distance of a first matching point pair is less than the boundary, it is determined to be a matching point pair with high spatial fit and reliable anatomical correspondence, and is retained. If the distance is greater than or equal to the boundary, it is determined that there may be slight anatomical deviation, such as POI shift caused by slight patient tremors, and is discarded. Finally, the processor defines the retained matching point pairs as second matching point pairs, which serve as the core data for subsequent transformation model selection of the image matching pair, ensuring that the matching point pairs input to the transformation model have both quantity and accuracy.

[0070] In this application, the Lowe ratio test is used to accurately eliminate erroneous matching pairs affected by interference points, reducing the number of low-precision matching pairs entering subsequent processes from the source. Distance sorting provides an ordered data foundation for distance distribution analysis, avoiding screening bias caused by disordered data. Method 2 (C5) is used to calculate the distance boundary point with the 0.9 quantile value, effectively eliminating discrete value interference, which is more in line with the actual distance distribution characteristics of CBCT bone imaging than the traditional screening method of multiplying the maximum value and the proportional coefficient. Finally, the screening further retains high-fit matching point pairs. These four progressive steps not only solve the problem of erroneous matching but also address the problem of discrete value interference, ultimately obtaining a sufficient number of reliable second matching point pairs. These high-quality matching point pairs provide accurate data support for the subsequent screening of the optimal transformation model, avoiding the deviation of transformation model parameters caused by inferior matching point pairs, thereby ensuring the spatial alignment accuracy between the second scanning area and the first scanning area, and laying a solid matching foundation for the overall accuracy of heavy-position CBCT imaging.

[0071] As an optional implementation, in step S13, combining matching point pairs, the optimal transformation model is selected layer by layer from multiple transformation models according to preset filtering rules, including: Step S131: Determine the transformation model associated with each image matching pair and the transformation rules of the transformation model; Step S132: Take the second matching point pair in the image matching pair that satisfies the corresponding transformation rule as the inlier. If the number of inliers is greater than the number threshold, retain the transformation model associated with the image matching pair and use it as the first transformation model. Step S133: Determine the transformation matrix corresponding to each first transformation model, and select the second transformation model whose rotation elements are greater than the rotation threshold from multiple transformation matrices; Step S134: Determine the product of the maximum interior point and the preset scaling parameter in the second transformation model, and select the third transformation model with a greater number of interior points than the product value; Step S135: Determine the matching image number corresponding to each third transformation model; determine the fourth transformation model associated with the number that appears most frequently; if the number of fourth transformation models exceeds one, select the transformation model associated with the candidate matching image closest to the center of the overlapping region as the optimal transformation model.

[0072] In step S131, for each image matching pair, the processor uses the RANSAC (Random Sample Consensus) algorithm to calculate the uniquely associated transformation model of the image matching pair. The RANSAC algorithm iteratively filters the spatial adjustment rules that enable the second matching point pairs in the image matching pair to achieve alignment, i.e., the transformation rules, which include parameters such as translation and rotation. At the same time, it counts the number of second matching point pairs that conform to the transformation rules and defines these matching point pairs that conform to the rules as interior points.

[0073] In step S132, the processor, based on the determined transformation rules, re-verifies the number of second matching point pairs (i.e., interior points) that satisfy the rules in each image matching pair, and uses a threshold number (e.g., 20) as a screening criterion. If the number of interior points corresponding to a certain transformation model is greater than the threshold number, it indicates that the model can achieve spatial alignment for most matching point pairs, conforming to the correspondence between the skeletal anatomical structures of the two regions. The processor retains this transformation model and defines it as the first transformation model. If the number of interior points is not greater than the threshold number, it is determined that the model cannot support accurate spatial alignment and is discarded. Through this step, transformation models with insufficient interior points and low reliability are further filtered out, narrowing the range of candidate models and reducing the computational load of subsequent screening.

[0074] In step S133, the processor first converts the transformation rules of each first transformation model into a corresponding transformation matrix. This matrix contains specific elements related to image rotation transformation (e.g., m1 and m5 in the matrix, which theoretically have a value of 1 when there is no rotation).

[0075] The transformation matrix is ​​shown below.

[0076]

[0077] In accordance with the clinical requirements of CBCT weight-bearing scans, patients need to remain as still as possible to avoid bone displacement. Therefore, the processor sets a rotation threshold, for example, 0.8, for these rotation-related elements. If any rotation-related element in the transformation matrix of a first transformation model is less than this rotation threshold, the model is considered to have a high probability of significant calculation error and cannot reflect the true spatial relationship of the bones, and is therefore excluded. Only first transformation models in which all rotation-related elements are not less than the rotation threshold are retained and defined as second transformation models. This screening process effectively eliminates models with rotational deviations caused by slight patient rotation by constraining rotation-related elements, ensuring that subsequent alignment conforms to the physiological position of the bones.

[0078] In step S134, the processor first counts the number of interior points corresponding to all second transformation models, finds the maximum number of interior points, and then calculates the product of the preset ratio parameter (e.g., 0.6) as a screening threshold. Subsequently, the processor compares the number of interior points of each second transformation model with this product value. If the number of interior points of a second transformation model is greater than the product value, it indicates that the model can align most matching point pairs, avoiding local alignment but overall misalignment (e.g., only a few matching points are aligned while most are offset). The processor retains this model and defines it as a third transformation model. If the number of interior points is not greater than the product value, it is determined that the model still has a large error and is discarded. This step, through the proportional constraint of the number of interior points, further improves the accuracy of candidate models, ensuring that the retained models fit the overall matching relationship.

[0079] In step S135, the processor first retrieves the matching image sequence number associated with each third transform model, counts the occurrence frequency of all sequences, finds the sequence number with the most occurrences (i.e., the mode sequence number), and defines the third transform model associated with this sequence number as the fourth transform model. A high occurrence frequency indicates that multiple image matching pairs support the candidate matching image corresponding to this sequence number, and their spatial correspondence is more reliable. If the number of fourth transform models exceeds one, that is, there are multiple sequences with the same and most occurrence frequencies, the processor further calculates the distance between the candidate matching images corresponding to these sequences and the center of the overlapping region, selects the transform model associated with the candidate matching image with the closest distance, and finally determines it as the optimal transform model. The image closest to the center of the overlapping region is selected because the skeletal anatomy at the center of the overlapping region is more stable and less affected by edge artifacts, which can further improve the reliability of the optimal model.

[0080] In this application, the RANSAC algorithm is used to calculate the transformation model, and a threshold for the number of inliers (e.g., 20) is used to eliminate inferior models that cannot fit most matching point pairs. This effectively avoids interference from a few erroneous matching points caused by slight motion and artifacts in CBCT scans, laying the foundation for model reliability. The number of inliers is verified a second time to further narrow down the range of low-reliability models and reduce the amount of subsequent screening calculations. The transformation matrix is ​​constrained by a threshold for rotational elements to exclude models with rotational deviations caused by slight patient rotations, ensuring that the model conforms to the physiological positional correspondence of the skeleton. The product of a preset proportional parameter and the maximum number of inliers is used as the screening threshold to eliminate models that are only locally aligned, avoiding the problem of a few matching pairs being aligned but the whole being misaligned. Matching images are selected based on the frequency of occurrence of the matching image sequence number. When multiple models compete, the model closest to the center of the overlapping area is selected to ensure that the final optimal transformation model has both high support and anatomical structural stability.

[0081] As an optional implementation, in step 105, based on the spatial correspondence, a fusion operation is performed on the main image region and the transformed matching image region to obtain the fused overlapping region, including: Step S21: Determine the overlap of overlapping images in the main image region and the transformed matching image region; Step S22: Configure an image index for each overlapping image, wherein the image index is used to identify the positional order of the overlapping image from the main image region to the transformed matching image region; Step S23: Determine the weighted curve based on the number of overlaps and the image index. The weighted curve is monotonically changing and continuously differentiable, which is used to achieve a smooth transition in the overlapping area. Step S24: Using the weighted curve as the weight, perform weighted fusion on the main image region and the transformed matching image region to obtain the fused overlapping region.

[0082] In step S21, the processor first locates overlapping images in the main image region and the transformed matching image region, i.e., frames that completely cover each other spatially and contain the same skeletal interest points. Then, the processor compares the spatial coordinates and interest points of the two regions and counts the total number of these overlapping frames, defining it as the overlap number N (e.g., if the main image region has 8 frames, the transformed matching image region has 8 frames, and the overlapping frames that completely cover each other and contain the same interest points are 6, then N=6). This process ensures that the subsequent weighted curve calculation has an accurate baseline, avoiding errors in weight allocation due to statistical deviations.

[0083] In step S22, the processor assigns a unique image index j to each overlapping image according to the spatial transition order from the main image region to the transformed matching image region. For example, when N=6, the indices j are 0, 1, 2, 3, 4, and 5, corresponding to the positional order from the main image region side to the matching image region side. The core function of this image index is to establish the correspondence between the overlapping image positions and weights, ensuring that the subsequent weighted curve can accurately allocate weights according to spatial order, and always based on the spatial consistency of the points of interest.

[0084] In step S23, the processor calculates the weight of each overlapping image based on the determined number of overlaps N and the image index j using a preset formula (where w is the weighted curve value): The formula for calculating the weighted curve is:

[0085] Where w is the weighted curve, which must be monotonically changing and continuously differentiable, and the weighted curve varies from 0 to 1. N is the number of overlapping images, and j is the image index.

[0086] In the above formula, the value of w decreases monotonically as the index j increases, matching the positional order from the main image region to the matching image region. When j = 0 (overlapping image on the main image region side), w is close to 1, meaning that the image weight is mainly based on the pixels of the main image region, and the spatial position of the interest point in the image is closer to the reference of the first scan region. When j = N-1 (overlapping image on the matching image region side), w is close to 0, meaning that the image weight is mainly based on the pixels of the transformed matching image region, while ensuring the spatial alignment of the interest point with the main image region. At the same time, this formula ensures that the change of w satisfies the continuous differentiability property, that is, the transition of the weight from 1 to 0 has no abrupt inflection point (e.g., when j changes from 2 to 3, the change in w is smoothly connected with the change in j from 3 to 4), avoiding visual discontinuities in the skeletal structure corresponding to the interest point in the fused image due to abrupt weight changes.

[0087] In step S24, the processor iterates through all overlapping images between the main image region and the transformed matching image region, and performs weighted fusion on the main image region and the transformed matching image region. The fusion formula is imgmerge=img1*w+img2*(1-w). Here, imgmerge is the fused overlapping region, img1 is the main image region, and img2 is the transformed matching image region. After the processor completes the calculation for all images, it integrates them to obtain a fused overlapping region with no obvious boundaries and spatial continuity of interest points. Figure 4 This is a schematic diagram of image weighted fusion.

[0088] In this application, the number of overlapping images N is counted based on points of interest to ensure accurate positioning of overlapping areas and avoid quantity deviations caused by misjudgment of non-critical areas, providing a reliable basis for weight calculation. Secondly, the image index j is combined with the matching degree of points of interest to ensure that the weight allocation is consistent with the logic of the skeletal spatial structure and prevent misalignment of position and weight. Furthermore, the monotonic change and continuous differentiability of the weighting curve, based on the spatial stability of points of interest, eliminates the discontinuity of skeletal details caused by abrupt weight changes, making the fusion transition more in line with the real skeletal anatomy. Finally, the weighting of the main image region and the transformed matching image region eliminates the brightness deviation caused by the difference in scanning dose.

[0089] Ultimately, the fused overlapping area can seamlessly connect with the non-overlapping areas of the first and second scan areas, forming an axial CBCT image with continuous points of interest and complete skeletal structure (such as the entire lower limb and the entire spine). This effectively avoids problems such as misjudgment of the skeletal force line deviation and microfractures corresponding to the points of interest by doctors due to fusion traces, providing precise image support for orthopedic clinical diagnosis and further ensuring the overall accuracy of weight-bearing CBCT imaging.

[0090] This application also provides a schematic diagram of the complete imaging process of weight-bearing CBCT, such as... Figure 5 As shown, it includes the following content.

[0091] Step B1: Select the main image area.

[0092] Based on the height information recorded during CBCT segmented scanning, the processor calculates the number of overlapping images between adjacent upper and lower scanning regions (the lower part is the first scanning region and the upper part is the second scanning region). Then, from the overlapping part of the first scanning region, it selects several images located at the center of the overlapping region (the specific number is determined based on clinical experience, such as 3-5 images to ensure regional stability) and defines this part of the images as the main image region.

[0093] Step B2: Select the matching image region.

[0094] The processor also selects several images from the overlapping part of the second scanning area that correspond to the spatial position of the main image area based on the scanning height information and the statistical results of the number of overlapping images, and delineates them as the matching image area.

[0095] Step B3: Select the main image to be selected.

[0096] The processor randomly selects one image from multiple images within the main image region as the candidate main image.

[0097] Step B4: Select the images to be matched.

[0098] The processor randomly selects one image from multiple images within the matching image region as a candidate matching image, which is used to match the candidate main image determined in step B3 to form an image matching pair.

[0099] Step B5: Matching the candidate main image and the candidate matching image.

[0100] The processor uses the candidate master image as a reference and performs image matching calculations between the candidate master image and the candidate matching image. For detailed matching procedures, please refer to [link / reference]. Figure 6 ( Figure 6 This is a detailed breakdown diagram of step B5, including sub-steps such as interest point detection, filtering, and transformation model calculation. The optimal transformation model is ultimately obtained, and this model uniquely corresponds to the index of the candidate matching image. Based on the optimal transformation model, the target matching image and the target main image are determined.

[0101] Step B6: Determine the spatial mapping relationship between the target matching image and the target main image.

[0102] The processor establishes a spatial location mapping relationship based on the coordinates of the matching points of interest between the target matching image and the target main image.

[0103] Step B7: Complete image transformation of the second scan area.

[0104] The processor applies the transformation model obtained in step B6 to all images in the second scanning area and performs spatial transformation operations. For example, according to the model parameters, it shifts all images in the second scanning area to the right by a preset distance (e.g., 3 mm) and rotates them clockwise by a preset angle (e.g., 0.2 degrees) to ensure that the overall spatial position of the second scanning area is aligned with that of the first scanning area, thus eliminating the misalignment caused by segmented scanning.

[0105] Step B8: Merge the main image region and the transformed matching image region.

[0106] The processor performs a weighted fusion operation on the image of the transformed matching image region and the image of the main image region: by gradually assigning weights (e.g., the weight of the candidate main image side decreases from 1 to 0.5, and the weight of the transformed matching image side increases from 0.5 to 1), the brightness discontinuity caused by the difference in scanning dose between the two regions is eliminated, and a smooth transition of the overlapping region is achieved.

[0107] Step B9: Image stitching complete.

[0108] The processor stitches together the non-overlapping area image of the first scan region, the fused overlapping area image obtained in step B8, and the non-overlapping area image of the second scan region in sequence along the axial direction (the vertical direction of the human body) to finally generate a complete CBCT image with a large range and no tomography in the axial direction (such as the lower limb from the ankle to the hip and the spine from the sacrum to the cervical spine), which meets the clinical needs for overall structural observation.

[0109] Figure 6 This is a schematic diagram of the matching process between the candidate main image and the candidate matching image, including the following content.

[0110] Step C1: Interest point detection.

[0111] For each image matching pair determined in steps B3 and B4, the processor uses the SIFT algorithm to detect points of interest in the image.

[0112] Step C2: Description of points of interest.

[0113] The processor then uses the SIFT algorithm to generate feature descriptions for each interest point detected in step C1. These descriptions include information such as the scale, orientation, and grayscale distribution of surrounding pixels, and can be used to accurately compare the consistency of interest points between the candidate main image and the candidate matching image.

[0114] Step C3: Point of interest matching.

[0115] Based on the feature description of interest points, the processor performs interest point matching on the candidate main image and the candidate matching image: for any interest point (denoted as the main point X) in each candidate main image, two interest points with the highest feature similarity are selected in the candidate matching image (denoted as the effective matching point A and the interference point B, respectively), forming the initial main point-matching point association relationship.

[0116] Step C4: Interest point matching pair filtering method 1 (obtain the first matching point pair).

[0117] The processor performs filtering on the initial associations formed in step C3: it calculates the distance (d1) between the principal point X and the valid matching point A, and the distance (d2) between the principal point X and the interference point B, and calculates the distance ratio (d1 / d2). If the distance ratio exceeds the empirical threshold of 0.7, the processor determines that the interference point B has too great an impact on the matching and the probability of accurate matching is low, and removes the matching pair from this association; if the distance ratio does not exceed 0.7, the processor determines that the valid matching point A has stronger feature consistency with the principal point X, retains the X+A matching pair, deletes the X+B matching pair, and finally obtains the first matching point pair after preliminary filtering.

[0118] Step C5: Interest point matching pair filtering method 2 (optimizing the first matching point pair).

[0119] The processor further performs precision filtering on the first matching point pairs obtained in step C4: it calculates the spatial distance between the two points of interest in each first matching point pair, statistically analyzes the distribution characteristics of all distances, and selects the 0.9 quantile (an empirical value, denoted as L) of the distance distribution. This quantile effectively eliminates extreme discrete values, such as erroneous matching point pair distances caused by artifacts. Subsequently, the processor removes first matching point pairs with distances exceeding 0.75*L, retaining first matching point pairs with closer distances and stronger spatial consistency, providing high-quality data for subsequent transformation model calculations.

[0120] Step C6: Transformation model calculation (RANSAC algorithm).

[0121] The processor employs the RANSAC algorithm to calculate a transformation model based on the first matching point pairs optimized in step C5. This transformation model consists of rules (including translation and rotation parameters) that enable spatial alignment of most first matching point pairs. Simultaneously, the processor counts the number of first matching point pairs that conform to these rules and defines these pairs as interior points. If the number of interior points is less than an empirical threshold (e.g., 20), the processor determines that the transformation model cannot fit the majority of matching relationships and has a significant deviation, and directly discards it; only transformation models with an interior point count not less than the threshold are retained.

[0122] Step C7: Transformation model screening method 3 (rotation constraint).

[0123] Because the first matching point pairs after step C5 screening may still have a small number of inaccurate associations, leading to deviations in the transformation model, the processor performs rotation constraint screening on the retained transformation models: in the transformation matrix corresponding to the transformation model, the two elements m1 and m5 are directly related to the image rotation transformation. Theoretically, if the patient remains stationary during the scan and the image does not rotate, m1 and m5 should be close to 1. Combining the clinical requirements of CBCT weight-bearing scans (the patient should be as still as possible), the processor sets an empirical threshold of 0.8. If m1 < 0.8 or m5 < 0.8 for a certain transformation model, it is determined that its calculation error probability is extremely high and it cannot reflect the true spatial relationship of the skeleton, and is therefore discarded.

[0124] Step C8: Transform model screening method 4 (interior point number ratio constraint).

[0125] The processor counts the number of interior points for all transformation models filtered in step C7, finds the maximum number of interior points, and then calculates 0.6 * the maximum number of interior points as the filtering threshold. If the number of interior points for a transformation model is less than this threshold, the processor determines that it can only achieve local alignment of the first matching point pair and overall misalignment (e.g., only 1-2 points are aligned, while most points are offset), and discards it; only transformation models with more interior points than the threshold are retained to ensure that the model can achieve overall spatial alignment between the two regions.

[0126] Step C9: Transformation model selection method 5 (determining the optimal transformation model).

[0127] The processor first retrieves each transformation model filtered in step C8 and associates it with its corresponding candidate matching image index. Then, it counts the occurrences of all indices and selects the index with the most occurrences (the mode index). The transformation model associated with this index is defined as a candidate transformation model. If there are more than one candidate transformation model (i.e., multiple indices have the same number of occurrences, all being the most frequent), the processor calculates the distance between the candidate matching images corresponding to these indices and the center of the overlapping region. It selects the transformation model associated with the candidate matching image closest to this distance, ultimately determining it as the optimal transformation model. The candidate matching image at the center of the overlapping region is least affected by edge artifacts, further improving model reliability.

[0128] The core process of this application revolves around multi-segment scanning, cyclic matching, precision selection, and seamless fusion, progressing step by step: First, by performing multiple segmented scans at different heights along the axial direction, multiple sets of local axial images covering the target area (such as the entire spine or lower limb) are obtained, laying the foundation for large-scale imaging; Second, cyclic matching processing is performed on the axial images of adjacent segments, combining multiple sets of candidate main images from the main image region with multiple sets of candidate matching images from the matching image region one by one to generate multiple pairs of matching images. By increasing the number of matching images, the reliability of detecting images at the same height position is improved, ensuring the reliability of the spatial correspondence between adjacent regions; Third, during the image matching process, five selection methods are applied sequentially. Methods (interest point matching pair screening method 1-2, transformation model screening method 3-5) are used to perform layer-by-layer verification and optimization on the initial interest point matching pairs and transformation models, effectively eliminating erroneous matching information and excluding transformation models with large deviations, thus significantly improving the matching accuracy and the accuracy of subsequent image transformations. Finally, in the image fusion stage, a specific fusion curve (such as a gradient cos weighted curve) is used to perform gradient fusion on the overlapping parts of adjacent regions after transformation and alignment. Through the continuous and smooth transition of weights between the main image region and the matching image region, seamless stitching of axial images is achieved, while eliminating color difference and artifacts, ultimately forming a large-scale, high-quality negative-weight CBCT imaging result in the axial direction.

[0129] This application can achieve the following beneficial effects.

[0130] 1. Overcoming the limitations of single-scan coverage and achieving large-area imaging: To address the issue of limited axial coverage in a single CBCT scan, this application effectively compensates for the insufficient coverage of a single scan by performing multi-segment scans at different heights and stitching together the obtained local images. This enables successful axial large-area CBCT imaging of the entire spine, lower limbs, and even the whole body, meeting the clinical needs for complete observation and evaluation of long-axial anatomical structures.

[0131] 2. Improve matching accuracy and reliability, and ensure spatial alignment quality: In the overlapping area of ​​adjacent scanning regions, this application performs an axial image cyclic matching process. By performing one-to-one matching calculations on multiple sets of candidate main images and candidate matching images, the number of matching samples is increased, the random error of matching a single set of images is reduced, the accuracy of spatial alignment of adjacent regions is improved, and the reliability of the matching results is enhanced, laying a reliable foundation for subsequent image transformation and fusion.

[0132] 3. Multi-round screening and optimization to reduce imaging defects: During the image matching process, this application applies five screening methods in sequence (interest point matching pair screening method 1-2, transformation model screening method 3-5) to optimize the initial interest point matching pairs and transformation models layer by layer, effectively improving the adaptability of the matched images and the accuracy of the transformation models. This avoids spatial misalignment problems that may occur in the image fusion stage from the source, and significantly reduces or even eliminates common stitching seams, color differences and artifacts after image stitching, ensuring the quality and stability of the final image.

[0133] 4. Gradual weighted fusion to enhance image continuity: This application uses a gradual cosine weighted curve to fuse overlapping image regions. By ensuring a smooth transition of weights between the main image region and the transformed matching image region, visual breaks caused by abrupt weight changes are avoided. This further reduces imaging defects (seams, color differences, artifacts) at the stitching interface, ensuring the continuity of anatomical structure and visual consistency of the stitched image in the axial direction, which meets the requirements of clinical diagnosis for image detail and overall coherence.

[0134] This application provides a complete imaging device for weight-bearing CBCT, such as... Figure 7 As shown, the device includes: The traversal module 701 is used to process all scanned regions obtained by segmented scanning of cone-beam computed tomography in a bottom-to-top order. The determining module 702 is used to determine a first scanning region located below and a second scanning region located above for two adjacent scanning regions processed sequentially, and to obtain the overlapping area between the first scanning region and the second scanning region, wherein the lower half of the overlapping area is the main image region and the upper half of the overlapping area is the matching image region. The selection module 703 is used to select the target main image and the target matching image with the highest matching degree from multiple candidate main images in the main image region and multiple candidate matching images in the matching image region, and to determine the spatial position correspondence. The image transformation module 704 is used to perform image transformation operations on all images in the second scanning area using the transformation model corresponding to the target matching image; The fusion module 705 is used to perform a fusion operation on the main image region and the transformed matching image region based on the spatial position correspondence to obtain the fused overlapping region. The stitching module 706 is used to stitch together the non-overlapping areas in the first scanning area, the fused overlapping areas, and the non-overlapping areas in the second scanning area in sequence to obtain a continuous scanning image in the axial direction.

[0135] Optionally, module 703 is selected for: Multiple candidate main images in the main image region and multiple candidate matching images in the matching image region are cyclically matched to form multiple image matching pairs, wherein each image matching pair is associated with a transformation model; For each image matching pair, perform interest point detection and matching filtering to determine multiple matching point pairs after filtering; By combining matching point pairs, the optimal transformation model is selected layer by layer from multiple transformation models according to preset filtering rules; Based on the matching image number associated with the optimal transformation model, the target matching image corresponding to the matching image number is determined. Each candidate matching image corresponds to a matching image number, and each matching image number is associated with multiple transformation models. The image located in the middle among multiple candidate main images is selected as the target main image.

[0136] Optionally, module 703 is specifically used for: Interest point detection and Lowe ratio test are performed on each image matching pair to obtain multiple first matching point pairs after preliminary screening; Based on the distance between the points of interest in each first matching point pair, sort the first matching point pairs in ascending order of distance; Based on the distance distribution of the first matching point pair after sorting, the distance boundary point is determined by combining the distance corresponding to the preset quantile value and the preset proportional coefficient. From the first pair of matching points, select the second pair of matching points whose distance to the point of interest is less than the distance boundary point, and use them as the filtered multiple pairs of matching points.

[0137] Optionally, module 703 is specifically used for: Determine the transformation model associated with each image matching pair and the transformation rules of the transformation model; The second matching point pair in the image matching pair that satisfies the corresponding transformation rule is taken as the inlier. If the number of inliers is greater than the number threshold, the transformation model associated with the image matching pair is retained and taken as the first transformation model. Determine the transformation matrix corresponding to each first transformation model, and select the second transformation model whose rotation elements are greater than the rotation threshold from multiple transformation matrices; Determine the product of the maximum interior point and the preset scaling parameter in the second transformation model, and select the third transformation model with a greater number of interior points than the product value; Based on the matching image sequence number corresponding to each third transformation model, the third transformation model corresponding to the sequence number that appears most frequently is taken as the optimal transformation model.

[0138] Optionally, module 703 is specifically used for: Determine the matching image number corresponding to each third transformation model; Determine the fourth transformation model associated with the most frequently occurring sequence number; If the number of fourth transformation models exceeds one, the transformation model associated with the candidate matching image closest to the center of the overlapping region is selected as the optimal transformation model.

[0139] Optionally, the fusion module 705 is specifically used for: Determine the amount of overlap between the overlapping images in the main image region and the transformed matching image region; Configure an image index for each overlapping image, where the image index is used to identify the positional order of the overlapping images from the main image region to the transformed matching image region; A weighted curve is determined based on the number of overlaps and the image index. The weighted curve is monotonically changing and continuously differentiable, which is used to achieve a smooth transition of the overlapping region. Using a weighted curve as the weight, the main image region and the transformed matching image region are weighted and fused to obtain the fused overlapping region.

[0140] Optionally, the formula for calculating the weighted curve is:

[0141] Where w is the weighted curve, N is the number of overlapping images, and j is the image index.

[0142] like Figure 8 As shown, this application provides an electronic device including a processor 801, a communication interface 802, a memory 803, and a communication bus 804, wherein the processor 801, the communication interface 802, and the memory 803 communicate with each other through the communication bus 804.

[0143] The memory 803 is used to store computer programs.

[0144] In one embodiment of this application, the processor 801, when executing the program stored in the memory 803, implements the complete imaging method of negative weight CBCT provided in any of the foregoing method embodiments.

[0145] This application also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the complete imaging method for weight-bearing CBCT as provided in any of the foregoing method embodiments.

[0146] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0147] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented using software plus a general-purpose hardware platform, or of course, using hardware. Based on this understanding, the above technical solutions, in essence or the parts that contribute to the related technology, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0148] It should be understood that the terminology used herein is for the purpose of describing particular exemplary embodiments only and is not intended to be limiting. Unless the context clearly indicates otherwise, the singular forms “a,” “an,” and “described” as used herein may also include the plural forms. The terms “comprising,” “including,” “containing,” and “having” are inclusive and therefore indicate the presence of the stated features, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, elements, components, and / or combinations thereof. The method steps, processes, and operations described herein are not construed as requiring them to be performed in a particular order described or illustrated unless the order of performance is explicitly indicated. It should also be understood that additional or alternative steps may be used.

[0149] The above description is merely a specific embodiment of this application, enabling those skilled in the art to understand or implement this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features claimed herein.

Claims

1. A complete imaging method for negative-position CBCT, characterized in that, The method includes: All scanned regions obtained from segmented scanning of cone-beam computed tomography are processed sequentially from bottom to top. For the two adjacent scanning regions processed sequentially, a first scanning region located below and a second scanning region located above are determined, and the overlapping region between the first scanning region and the second scanning region is obtained, wherein the lower half of the overlapping region is the main image region and the upper half of the overlapping region is the matching image region; From multiple candidate main images in the main image region and multiple candidate matching images in the matching image region, select the target main image and the target matching image with the highest matching degree and determine the spatial position correspondence. Using the transformation model corresponding to the target matching image, image transformation operations are performed on all images in the second scanning region; Based on the spatial correspondence, a fusion operation is performed on the main image region and the transformed matching image region to obtain the fused overlapping region; The non-overlapping areas in the first scanning area, the fused overlapping area, and the non-overlapping areas in the second scanning area are sequentially stitched together to obtain a continuous scanning image in the axial direction.

2. The method according to claim 1, characterized in that, Selecting the target main image and the target matching image with the highest matching degree from multiple candidate main images in the main image region and multiple candidate matching images in the matching image region includes: Multiple candidate main images in the main image region and multiple candidate matching images in the matching image region are cyclically matched to form multiple image matching pairs, wherein each image matching pair is associated with a transformation model; For each image matching pair, interest point detection and matching filtering are performed to determine multiple filtered matching point pairs; Based on the matching point pairs, the optimal transformation model is selected layer by layer from multiple transformation models according to the preset filtering rules; Based on the matching image number associated with the optimal transformation model, the target matching image corresponding to the matching image number is determined, wherein each candidate matching image corresponds to a matching image number, and each matching image number is associated with multiple transformation models. The main image located in the middle position among multiple candidate main images is taken as the target main image.

3. The method according to claim 2, characterized in that, For each image matching pair, interest point detection and matching filtering are performed to determine multiple filtered matching point pairs, including: For each image matching pair, interest point detection and Lowe ratio test are performed to obtain multiple first matching point pairs after preliminary screening; Based on the distance between the points of interest in each first matching point pair, sort the first matching point pairs in ascending order of distance; Based on the distance distribution of the first matching point pair after sorting, the distance boundary point is determined by combining the distance corresponding to the preset quantile value and the preset proportional coefficient. From the first pair of matching points, select a second pair of matching points whose distance to the point of interest is less than the distance boundary point, and use them as the filtered multiple pairs of matching points.

4. The method according to claim 2, characterized in that, Combining the matching point pairs, the optimal transformation model is selected layer by layer from multiple transformation models according to preset filtering rules, including: Determine the transformation model associated with each image matching pair and the transformation rules of the transformation model; The second matching point pair in the image matching pair that satisfies the corresponding transformation rule is taken as the inlier. If the number of the inlier is greater than the number threshold, the transformation model associated with the image matching pair is retained and taken as the first transformation model. Determine the transformation matrix corresponding to each first transformation model, and select the second transformation model whose rotation elements are greater than the rotation threshold from multiple transformation matrices; Determine the product of the maximum interior point and the preset scaling parameter in the second transformation model, and filter out the third transformation model with a greater number of interior points than the product value; Based on the matching image sequence number corresponding to each third transformation model, the third transformation model corresponding to the sequence number that appears most frequently is taken as the optimal transformation model.

5. The method according to claim 4, characterized in that, Based on the matching image sequence number corresponding to each third transform model, the third transform model corresponding to the sequence number that appears most frequently is selected as the optimal transform model, including: Determine the matching image number corresponding to each third transformation model; Determine the fourth transformation model associated with the most frequently occurring sequence number; If the number of fourth transformation models exceeds one, the transformation model associated with the candidate matching image closest to the center of the overlapping region is selected as the optimal transformation model.

6. The method according to claim 1, characterized in that, Based on the spatial correspondence, a fusion operation is performed on the main image region and the transformed matching image region to obtain the fused overlapping region, which includes: Determine the number of overlaps between the main image region and the transformed matching image region; An image index is configured for each of the overlapping images, wherein the image index is used to identify the positional order of the overlapping images from the main image region to the transformed matching image region; A weighted curve is determined based on the number of overlaps and the image index, wherein the weighted curve is monotonically changing and continuously differentiable, and is used to achieve a smooth transition of the overlapping region; Using the weighted curve as the weight, the main image region and the transformed matching image region are weighted and fused to obtain the fused overlapping region.

7. The method according to claim 6, characterized in that, The formula for calculating the weighted curve is: Where w is the weighted curve, N is the number of overlapping images, and j is the image index.

8. A complete imaging device for heavy-duty CBCT, characterized in that, The device includes: The traversal module is used to process all scanned regions obtained by segmented scanning of cone-beam computed tomography in a bottom-up order. The determining module is used to determine a first scanning region located below and a second scanning region located above for two adjacent scanning regions processed sequentially, and to obtain the overlapping area between the first scanning region and the second scanning region, wherein the lower half of the overlapping area is the main image region and the upper half of the overlapping area is the matching image region; The selection module is used to select the target main image and the target matching image with the highest matching degree from multiple candidate main images in the main image region and multiple candidate matching images in the matching image region, and to determine the spatial position correspondence. The image transformation module is used to perform image transformation operations on all images in the second scanning area using the transformation model corresponding to the target matching image; The fusion module is used to perform a fusion operation on the main image region and the transformed matching image region based on the spatial position correspondence to obtain the fused overlapping region. The stitching module is used to sequentially stitch together the non-overlapping areas in the first scanning area, the fused overlapping area, and the non-overlapping areas in the second scanning area to obtain a continuous scanning image in the axial direction.

9. An electronic device, characterized in that, It includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; Memory, used to store computer programs; A processor, when executing a program stored in memory, implements the method described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the method described in any one of claims 1-7.

Images