Automatic Detection and Comparison Method and System for Lung Nodules on Chest CT Images

Abstract

This invention provides an automatic detection and comparison method and system for lung nodules in chest CT images, relating to the field of data processing technology. The method includes: determining the sub-marker region into which the marked nodule location falls within the initial state sub-marker group; mapping the marked nodule location to the current CT image based on the vertex correspondence between the initial state and current state sub-marker regions to obtain a predicted matching location; correcting the predicted matching location along the centroid direction of the sub-marker region using the local deformation coefficient corresponding to the sub-marker region to obtain a corrected nodule location; constructing a spherical search area centered on the corrected nodule location and with the diameter of the marked nodule as the radius; and performing nodule detection on the current CT image within the spherical search area to complete the automatic detection and comparison of lung nodules. This invention improves the matching accuracy of automatic detection and comparison of lung nodules between two consecutive CT images.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to an automatic detection and comparison method and system for lung nodules in chest CT images. Background Technology

[0002] In the follow-up comparison of pulmonary nodules in the same subject based on chest CT images, it is often necessary to map the locations of nodules detected in previous CT images to the current CT images so that nodules can be automatically searched and compared in the corresponding areas of the current images. This mapping process mostly relies on the registration between images of different time phases. Since the respiratory depth, body position and other conditions of the subject may be different between the two examinations, the deformation of lung tissue in different anatomical regions often varies. For example, the respiratory displacement amplitude and deformation pattern of lung tissue located near the tracheal carina, around the hilum and in the subpleural region are not completely consistent. Some existing methods use a global deformation field based on the whole lung contour or sparse anatomical landmarks for position transfer, which sometimes fails to fully reflect the non-uniformity of the degree of deformation in local areas.

[0003] Such local deformation differences may cause the predicted location of nodules obtained from previous image mapping to deviate from the actual corresponding anatomical location in the current image. If nodule detection and comparison are performed directly with the predicted location as the center, and the search area size is not properly selected, in some cases the search radius needs to be expanded to cover the deviation, which may introduce more false positive candidates and reduce the comparison efficiency. If the search range is too conservative, it may miss nodules that have shifted, affecting the accuracy of automatic detection and comparison. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide an automatic detection and comparison method and system for lung nodules in chest CT images, which improves the matching accuracy of automatic detection and comparison of lung nodules between two consecutive CT images.

[0005] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows: Firstly, an automatic detection and comparison method for lung nodules on chest CT images, the method comprising: Step 1: Obtain the current CT image and the previous CT image of the same subject. The previous CT image includes the marked location and diameter of the nodules. Step 2: Extract the tracheal carina, the bifurcation of the right upper lobe bronchus, and the upper margin of the left pulmonary artery arch from the lung parenchyma region of the previous CT image as three anatomical landmarks. Step 3: Construct the initial three-boundary region based on the coordinates of the three anatomical landmarks in the previous CT image. Divide the initial three-boundary region by connecting the centroid of the initial three-boundary region to the three sides of the initial three-boundary region to obtain the initial sub-boundary group. Step 4: Construct the current three-boundary region based on the coordinates of the three anatomical landmarks in the current CT image. Divide the current three-boundary region by connecting the centroid of the current three-boundary region to the three sides of the current three-boundary region to obtain the current sub-boundary group. Step 5: Calculate the area ratio of the corresponding sub-marker regions in the initial state sub-marker group and the current state sub-marker group to obtain the local deformation coefficient; Step 6: Determine the sub-marker region in which the marked nodule location falls within the initial state sub-marker group; based on the vertex correspondence between the initial state three-marker regions and the current state three-marker regions, map the marked nodule location to the current CT image to obtain the predicted matching location; along the centroid direction of the sub-marker region, correct the predicted matching location using the local deformation coefficient corresponding to the sub-marker region to obtain the corrected nodule location; Step 7: Construct a spherical search area with the corrected nodule location as the center and the marked nodule diameter as the radius. Perform nodule detection on the current CT image within the spherical search area to complete the automatic detection and comparison of lung nodules.

[0006] Secondly, the automatic detection and comparison system for lung nodules on chest CT images includes: The acquisition module is used to acquire current and previous CT images of the same subject. The previous CT images include the marked nodule locations and marked nodule diameters. The extraction module is used to extract the tracheal carina, the bifurcation of the right upper lobe bronchus, and the upper margin of the left pulmonary artery arch from the lung parenchyma region of the previous CT image as three anatomical landmarks. The construction module is used to construct the initial three-boundary region based on the coordinates of the three anatomical landmarks in the previous CT image. The initial three-boundary region is divided by connecting the centroid of the initial three-boundary region to the three sides of the initial three-boundary region to obtain the initial sub-boundary group. The segmentation module is used to construct the current three-boundary region based on the coordinates of the three anatomical landmarks in the current CT image. The current three-boundary region is segmented by connecting the centroid of the current three-boundary region to the three sides of the current three-boundary region to obtain the current sub-boundary group. The calculation module is used to calculate the area ratio of the corresponding sub-marker regions in the initial state sub-marker group and the current state sub-marker group, and obtain the local deformation coefficient. The correction module is used to determine the sub-marker region in which the marked nodule location falls within the initial state sub-marker group; based on the vertex correspondence between the initial state three-marker region and the current state three-marker region, the marked nodule location is mapped to the current CT image to obtain the predicted matching location; along the centroid direction of the sub-marker region, the predicted matching location is corrected using the local deformation coefficient corresponding to the sub-marker region to obtain the corrected nodule location; The detection module is used to construct a spherical search area with the corrected nodule position as the center and the labeled nodule diameter as the radius. Within the spherical search area, nodule detection is performed on the current CT image to complete the automatic detection and comparison of lung nodules.

[0007] Thirdly, a computing device includes: One or more processors; A storage device for storing one or more programs that, when executed by one or more processors, cause the one or more processors to implement the method.

[0008] Fourthly, a computer-readable storage medium storing a program that, when executed by a processor, implements the method.

[0009] The above-described solution of the present invention has at least the following beneficial effects: The initial mapping of nodule positions is achieved by establishing the correspondence between the vertices of the three boundary marker areas. A secondary correction is then performed on the predicted matching positions using the local deformation coefficient and offset direction. This effectively mitigates the positional bias caused by uneven local deformation in traditional mapping methods, allowing the corrected nodule positions to better match the actual anatomical locations in the current CT image, thus improving the accuracy of nodule position mapping. A spherical search area is constructed based on the corrected nodule positions and the diameter of the previously labeled nodules, allowing the search range to better reflect the actual displacement of the nodules. This scheme reasonably constrains the nodule search interval, reducing the number of false positive candidate nodules introduced by an excessively large search range, thus improving the overall efficiency of lung nodule follow-up comparison and contributing to the stability and reliability of automated lung nodule follow-up detection. Attached Figure Description

[0010] Figure 1 This is a schematic flowchart of an automatic detection and comparison method for lung nodules in chest CT images provided by an embodiment of the present invention.

[0011] Figure 2 This is a schematic diagram of an automatic lung nodule detection and comparison system for chest CT images provided in an embodiment of the present invention. Detailed Implementation

[0012] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0013] like Figure 1 As shown, embodiments of the present invention propose an automatic detection and comparison method for lung nodules in chest CT images, the method comprising the following steps: Step 1: Obtain the current CT image and the previous CT image of the same subject. The previous CT image includes the marked location and diameter of the nodules. Step 2: Extract the tracheal carina, the bifurcation of the right upper lobe bronchus, and the upper margin of the left pulmonary artery arch from the lung parenchyma region of the previous CT image as three anatomical landmarks. Step 3: Construct the initial three-boundary region based on the coordinates of the three anatomical landmarks in the previous CT image. Divide the initial three-boundary region by connecting the centroid of the initial three-boundary region to the three sides of the initial three-boundary region to obtain the initial sub-boundary group. Step 4: Construct the current three-boundary region based on the coordinates of the three anatomical landmarks in the current CT image. Divide the current three-boundary region by connecting the centroid of the current three-boundary region to the three sides of the current three-boundary region to obtain the current sub-boundary group. Step 5: Calculate the area ratio of the corresponding sub-marker regions in the initial state sub-marker group and the current state sub-marker group to obtain the local deformation coefficient; Step 6: Determine the sub-marker region in which the marked nodule location falls within the initial state sub-marker group; based on the vertex correspondence between the initial state three-marker regions and the current state three-marker regions, map the marked nodule location to the current CT image to obtain the predicted matching location; along the centroid direction of the sub-marker region, correct the predicted matching location using the local deformation coefficient corresponding to the sub-marker region to obtain the corrected nodule location; Step 7: Construct a spherical search area with the corrected nodule location as the center and the marked nodule diameter as the radius. Perform nodule detection on the current CT image within the spherical search area to complete the automatic detection and comparison of lung nodules.

[0014] In this embodiment of the invention, the initial mapping of nodule positions is completed through the correspondence of the vertices of the three boundary marker areas. Then, the predicted matching positions are corrected a second time by combining the local deformation coefficient and the offset direction. This can effectively weaken the positional deviation caused by uneven local deformation in traditional mapping methods, making the corrected nodule positions more consistent with the actual anatomical positions in the current CT image, thus improving the accuracy of nodule position mapping. Based on the corrected nodule positions and the diameter of the previously marked nodules, a spherical search area is constructed, which makes the search range more consistent with the actual displacement of the nodules. This scheme can reasonably constrain the nodule search interval, which can reduce the large number of false positive candidate nodules introduced by the excessively large search range, help improve the overall efficiency of lung nodule follow-up comparison, and help improve the stability and reliability of automatic lung nodule follow-up detection.

[0015] In a preferred embodiment of the present invention, step 1 involves acquiring the current CT image and the previous CT image of the same subject. The previous CT image includes the marked nodule location and the marked nodule diameter. Specifically, this includes retrieving the current chest CT image sequence and the previous chest CT image sequence of the same subject from the hospital's image archiving and communication system, and simultaneously reading the annotation information corresponding to the previous CT image. The annotation information includes the three-dimensional spatial coordinates of the lung nodules that have been marked in the previous examination, as well as the diameter value of the nodule. Then, the two images are preprocessed to resample the slice thickness of the two images to the same standard resolution, and the grayscale values ​​of the images are normalized to eliminate grayscale differences caused by different scanning devices.

[0016] This embodiment effectively mitigates the differences in image spatial scale and grayscale deviation caused by different scanning equipment, scanning parameters, and exposure conditions by uniformly performing grayscale normalization preprocessing on two consecutive chest CT images, so that the current image and the previous image are kept under similar standardized data dimensions; at the same time, the three-dimensional coordinates and diameter annotation information of the lung nodules in the previous image are completely preserved, reducing the calculation deviation caused by inconsistent data standards.

[0017] In a preferred embodiment of the present invention, step 2, extracting the tracheal carina, the bifurcation of the right upper lobe bronchus, and the upper margin of the left pulmonary artery arch as three anatomical landmarks within the lung parenchyma region of the previous CT image, may include: Step 201: Based on the density difference between lung tissue and surrounding tissues in the previous CT image, the lung parenchyma region is segmented using a density threshold. Based on the density difference between the tracheal wall and the air within the tracheal lumen in the lung parenchyma region, the tracheal tree structure is extracted. Specifically, this includes: distinguishing different tissues based on the standard HU value density characteristics of human chest CT images; the HU value range for air-containing tissues in the lung parenchyma is fixed as a negative value range; the HU values ​​of soft tissues such as chest wall muscles and bones, as well as hard tissues, are higher than those of the lung parenchyma. A density threshold range of -950HU to -700HU is selected for the lung parenchyma. All pixels in the previous CT image are traversed, and the HU value of each pixel is compared one by one. All pixel regions with HU values ​​within this threshold range are selected and retained, while the remaining pixels are... After removing all primitive regions, a preliminary discrete distribution of candidate lung parenchyma pixel regions is obtained. Next, connected component analysis is performed on the initially selected candidate lung parenchyma regions. The connected component analysis uses an eight-neighbor rule. The specific determination and execution process is as follows: taking each pixel to be determined within the candidate region as the center, the pixels adjacent to that pixel in eight directions (up / down, left / right, top-left, top-right, bottom-left, and bottom-right) are checked simultaneously. The grayscale and region affiliation attributes of the eight adjacent pixels are compared one by one. If both the adjacent pixels and the center pixel belong to the selected candidate lung parenchyma pixels, then the two types of pixels are determined to belong to the same connected region. This process is repeated for all candidate pixels to complete the classification of all regions, dividing all non-adjacent pixel regions that cannot be connected by the eight-neighbor relationship into independent connected regions.

[0018] After dividing all connected components, the total number of pixels contained in each independent connected component is counted. The total number of pixels represents the overall volume of the connected component. The two independent connected components with the largest volume values ​​are selected and retained, corresponding to the left and right lung regions of the human body, respectively. These two connected components are determined as the initial regions of the left and right lungs, and all other interference connected components with too small a volume are removed. Morphological closing operations are performed on the obtained initial regions of the left and right lungs. The morphological closing operation uniformly adopts a 5×5 rectangular structure operator. First, the lung parenchyma region is expanded to fill the small gaps, gaps and defects caused by the image scanning at the edge of the lung parenchyma. Then, the expanded region is eroded to restore the original overall outline size of the lung parenchyma. After the closing operation, the complete lung parenchyma region is obtained, and the accurate extraction of lung parenchyma is completed. Within the already extracted complete lung parenchyma region, only the lungs are targeted. A secondary threshold screening is performed on the image area covered by the lung parenchyma. A density threshold is set based on the density difference between the soft tissue of the tracheal wall and the air in the tracheal lumen. The HU value range of the pure air area in the tracheal lumen is -1020HU to -960HU, and the HU value of the soft tissue of the tracheal wall is significantly higher than this range. Selecting this air density threshold range, all pixels in the lung parenchyma area are traversed, and all pixels with HU values ​​within this threshold range are selected to obtain all air distribution areas of the tracheal lumen in the lung parenchyma. The selected air areas of the tracheal lumen are then subjected to eight-neighbor connectivity analysis. Specifically, the regions are classified according to the aforementioned complete eight-neighbor pixel determination rules. Along the longitudinal extension direction of the human trachea, all spatially continuous and coherent air connectivity areas are integrated, and scattered small air-containing interference areas in the lung parenchyma are removed. All effectively connected air areas of the tracheal lumen are completely spliced ​​and integrated to reconstruct the tracheal tree structure.

[0019] Step 202: Based on the tracheal tree structure, tracheal cross-sections are extracted layer by layer along the head-to-foot direction to obtain a tracheal cross-section sequence. Based on the change in the number of tracheal cross-sections at each level in the sequence, the first level where the number of tracheal cross-sections changes from one to two is identified as the carina. Specifically, this includes: defining the head-to-foot direction as the top-to-bottom scanning direction; using the pre-processed uniform slice thickness as a standard, layer-by-layer slicing is performed along this fixed direction on all CT image layers covered by the tracheal tree structure, with a tracheal cross-section extracted separately for each independent image layer; all the layer-by-layer extracted tracheal cross-sections are arranged in an orderly manner according to the layer arrangement from head to foot to form a tracheal cross-section sequence; after completing the cross-section sequence construction, the entire tracheal cross-section sequence is traversed level by level from top to bottom. The tracheal lumen connectivity region within each independent layer is statistically analyzed. An eight-neighbor connectivity rule is used to distinguish individual independent tracheal lumen cross-sections within a layer, and the total number of independent tracheal lumen cross-sections within a single layer is counted one by one. The changes in the number of tracheal lumen cross-sections in adjacent layers are continuously compared layer by layer. The number of cross-sections in the upper adjacent layers is counted against the number of cross-sections in the current layer. When a sudden change in the number of cross-sections is detected between two consecutive adjacent layers, and the number of independent tracheal lumen cross-sections in all upper adjacent layers remains stable at one, while the number of independent tracheal lumen cross-sections in the current layer changes to two for the first time, the current layer is immediately determined to be the characteristic layer where the carina is located. After accurately locating the characteristic layer, the three-dimensional spatial coordinate parameters corresponding to the layer image are read, and the three-dimensional coordinate information of the center position of the layer is completely preserved, thereby confirming the location of the carina.

[0020] Step 203: Based on the tracheal lumen cross-section on the right side of the plane where the carina is located, determine the starting cross-section of the right main bronchus; based on the tracheal lumen cross-section on the left side of the plane where the carina is located, determine the starting cross-section of the left main bronchus; based on the starting cross-section of the right main bronchus, trace the outline of the right main bronchus cross-section layer by layer along the head-to-foot direction, and determine the plane where the first branch opening appears on the outer wall of the right main bronchus cross-section as the bifurcation point of the right upper lobe bronchus. Specifically, this includes: retrieving and locking the characteristic plane corresponding to the determined carina, reading the data within that plane after eight-neighbor connectivity analysis... Two independent tracheal cavity cross-sectional regions were identified. Based on the standard anatomical orientation of the human chest, the left and right regions were distinguished. Using the midsagittal line of the imaging plane as the dividing line, the independent tracheal cavity cross-section within the right-side spatial region of the human body was identified and determined as the starting cross-section of the right main bronchus, and the independent tracheal cavity cross-section within the left-side spatial region of the human body was identified and determined as the starting cross-section of the left main bronchus. After completing the initial cross-section positioning, the starting reference position of the right main bronchus cross-section was used as the tracking starting point. The scanning and tracking were performed layer by layer along the head-to-foot direction towards the feet, with the tracking process always confined to the extracted... Within the complete lung parenchyma and tracheal tree structure, the right main bronchus lumen region is completely extracted at each standard slice thickness of the imaging layer. For the right main bronchus lumen region extracted at each layer, the complete bronchial contour is defined using the eight-neighbor connectivity rule. The complete pixel contour morphology, boundary range, and closure status of the lateral wall of the right main bronchus at each layer are recorded throughout the process, continuously determining whether the lateral wall contour maintains a complete and closed tubular structure. During the iterative tracking process layer by layer, the contour integrity of the lateral wall of the bronchus at each layer is continuously compared. When the first detection is made... When a local gap or broken opening structure appears on the lateral wall contour of the right main bronchus in a certain imaging layer, and this opening structure extends outward to a new independent air-containing connecting area, and this connecting area conforms to the bronchial course characteristics and is not within the lumen of the main bronchus, this layer is determined to be the characteristic layer where the right main bronchus first shows a branch opening, and this layer is identified as the layer where the right upper lobe bronchus bifurcates. Finally, the overall three-dimensional spatial coordinates of this bifurcation characteristic layer are accurately read, and the three-dimensional coordinate information corresponding to the center position of the bifurcation opening is retained, thereby determining the bifurcation position of the right upper lobe bronchus.

[0021] Step 204: Based on the vascular morphology of the left hilum in the lung parenchyma, extract the course of the left pulmonary artery and determine the upper edge of the arched vessel above the left main bronchus as the upper edge of the left pulmonary artery arch. Based on the tracheal carina, the bifurcation of the right upper lobe bronchus, and the upper edge of the left pulmonary artery arch, obtain three anatomical landmarks, specifically: First, accurately locate the target level of the left hilum. Based on the distribution characteristics of human thoracic anatomy, the left hilum region corresponds to the imaging area below the tracheal carina and in the middle of the medial side of the left lung. Using the determined tracheal carina characteristic level as the starting reference, continuously traverse the imaging levels layer by layer along the head-to-foot direction to lock the core area at the junction of the left lung tissue and mediastinal tissue. This area is where pulmonary vessels and bronchial structures are concentrated, which is the continuous imaging level area corresponding to the left hilum.

[0022] A complete extraction of the left pulmonary artery's course was performed. Based on the density differences between blood vessels and lung and bronchial tissues in chest CT images, a HU value density range corresponding to the pulmonary artery tissue was defined. All pixels within the left hilar region were traversed, and all pixel regions matching the blood vessel density characteristics were selected to obtain candidate regions for the left hilar blood vessels. These candidate regions were then categorized using an eight-neighbor connectivity rule, eliminating scattered small vascular noise regions and retaining continuous, well-formed main trunk blood vessel connectivity regions. Combined with the anatomical distribution patterns of blood vessels in the lungs, the main trunk blood vessel region corresponding to the left pulmonary artery was identified and selected. Blood vessel regions at each level were connected layer by layer to completely fit and extract the overall course of the left pulmonary artery. The spatial position of the blood vessels was traced segment by segment along the predetermined course of the left pulmonary artery to accurately verify its location. The spatial relationship between the pulmonary artery and the left main bronchus was determined, locating the segment where the left pulmonary artery transversely crosses and covers the lumen of the left main bronchus. The morphological characteristics of the blood vessels in this segment were observed; the segment where the left pulmonary artery crosses the left main bronchus naturally forms an upwardly bulging arc-shaped vascular structure, distinguishing it from segments where the blood vessels run straight. After accurately locating this arc-shaped vascular structure, all edge pixels of the arc-shaped vascular structure were traversed, and the edge pixel with the highest overall spatial position was selected and determined as the upper edge of the left pulmonary artery arch. Finally, the three-dimensional spatial coordinate parameters corresponding to this feature point were accurately read and stored, completing the location of the upper edge of the left pulmonary artery arch. Thus, the accurate extraction and coordinate recording of three stable anatomical landmarks—the tracheal carina, the bifurcation of the right upper lobe bronchus, and the upper edge of the left pulmonary artery arch—were completed sequentially.

[0023] In this embodiment, three anatomical landmarks are located in the center of the lung, the right lung field, and the left hilar region, respectively. The spatial distribution is relatively balanced, which can help to construct a benchmark spatial system that can cover areas with high incidence of pulmonary nodules. This can improve the problem of positioning deviation and insufficient benchmark stability that are easy to occur with a single landmark.

[0024] In a preferred embodiment of the present invention, step 3, constructing an initial three-boundary region based on the coordinates of the three anatomical landmarks in the previous CT image, and dividing the initial three-boundary region by connecting the centroid of the initial three-boundary region to the three sides of the initial three-boundary region to obtain an initial sub-marker group, may include: Step 301: Based on the three-dimensional coordinates of the three anatomical landmarks in the previous CT images, connect the three anatomical landmarks pairwise with straight lines to obtain the initial three-landmark area enclosed by three sides. Based on the three sides of the initial three-landmark area, draw a line connecting the midpoint of each side to the opposite vertex, and determine the centroid of the initial three-landmark area by the intersection of the three lines. Specifically, this includes: reading the complete three-dimensional spatial coordinate data of the three anatomical landmarks (carina, right upper lobe bronchus bifurcation, and upper edge of left pulmonary artery arch) in the previous standardized CT images. Each anatomical landmark corresponds to a unique set of three-dimensional coordinate parameters. Select the three sets of anatomical landmark spatial points in sequence and pair them in pairs. Perform straight line fitting and connection processing on each pair of spatial points to generate three spatial line segments. The three spatial line segments are arranged in a closed loop in a head-to-tail manner to enclose a complete, closed-boundary, and spatially fixed area. A defined three-dimensional enclosed spatial region is defined as the initial three-boundary marker region, constructed based on three anatomical landmarks. After the initial three-boundary marker region is constructed, the three outer enclosing edges of the region are fully identified and extracted. For each enclosing edge, the spatial midpoint is located. For a single spatial edge, the three-dimensional coordinates of the two endpoints are read, and the unique spatial midpoint coordinates of the edge are calculated by averaging the spatial coordinates of the two endpoints. The precise location of all midpoints of the three enclosing edges is completed sequentially. For each enclosing edge, the corresponding diagonal anatomical landmark vertex is determined. This diagonal vertex is the remaining anatomical landmark point in the initial three-boundary marker region that does not participate in the current edge formation. The precise midpoint of each edge is connected to the corresponding diagonal vertex with a straight line in space, thereby generating three independent spatial lines distributed within the initial three-boundary marker region. The three internal spatial lines extend continuously within the initial three-boundary marker area and eventually converge at the same spatial point. The common overlapping point of the three spatial lines is the centroid of the initial three-boundary marker area.

[0025] Step 302: Connect the centroid of the initial state three-boundary marker region to the three vertices of the initial state three-boundary marker region to obtain three initial state dividing lines; based on the three initial state dividing lines, divide the initial state three-boundary marker region into three initial state sub-marker regions to obtain an initial state sub-marker group. Specifically, this includes: retrieving the three-dimensional coordinates of the centroid of the initial state three-boundary marker region, and simultaneously retrieving the three-dimensional spatial coordinates corresponding to the three anatomical marker vertices constituting the initial state three-boundary marker region; using the centroid of the initial state three-boundary marker region as a unified starting point, perform spatial straight line connection operations with each of the three anatomical marker vertices one by one to generate three directional lines in sequence. The three spatial straight lines that are uniform, regularly distributed, and do not intersect each other are the initial state dividing lines. These three initial state dividing lines extend from the central centroid to the surrounding vertices, completely penetrating the internal space of the initial state three boundary marker areas, uniformly dividing the overall closed space area of ​​the initial state three boundary marker areas into three closed sub-space areas. Each independent closed sub-space area obtained after uniform division is an initial state sub-boundary marker area adapted for local lung tissue deformation analysis. All three initial state sub-boundary marker areas are integrated and combined to form a complete initial state sub-boundary marker group.

[0026] In this embodiment, a closed three-marker region of the previous image is constructed using three stable anatomical landmarks. The overall reference region is divided into three independent sub-marker regions by finding the centroid by connecting the midpoints of the boundary lines and dividing the sub-regions based on the centroid. Each sub-region can correspond to a different local lung anatomical unit, which can better distinguish the spatial distribution differences of non-uniform lung deformation.

[0027] In a preferred embodiment of the present invention, step 4, constructing a current three-marker region based on the coordinates of the three anatomical landmarks in the current CT image, and dividing the current three-marker region by connecting the centroid of the current three-marker region to the three sides of the current three-marker region to obtain a current sub-marker group, may include: Step 401: Based on the three-dimensional coordinates of the three anatomical landmarks in the current CT image, connect the three anatomical landmarks pairwise with straight lines to obtain the current three-landmark region enclosed by three sides. Specifically, this includes: reusing the complete standardized processing flow and judgment rules for extracting anatomical landmarks from the previous CT image, performing the same anatomical landmark extraction operation on the preprocessed standardized CT image of the current image, and accurately locating and identifying three sets of fixed anatomical landmarks in sequence: the carina, the bifurcation of the right upper lobe bronchus, and the upper edge of the left pulmonary artery arch, ensuring that the landmark type, anatomical location, and extraction logic extracted in the current image are completely consistent with the previous image; accurately reading the complete three-dimensional spatial coordinate data corresponding to the three sets of anatomical landmarks in the coordinate system of the current CT image, with each anatomical landmark corresponding to a unique and independent set of three-dimensional coordinate parameters, and completely retaining the three sets of coordinate data for region construction; sequentially selecting the spatial points of the three sets of anatomical landmarks for pairwise pairing, and performing a spatial straight line fitting connection operation on each paired spatial point to sequentially generate three independent spatial straight line segments. The three spatial straight segments are arranged in a closed loop by connecting their beginning and end points in sequence, so that the three straight segments are connected and enclosed to form a three-dimensional closed spatial region with complete closed boundaries, fixed spatial range, and no gaps or gaps. The closed spatial region enclosed by the current image anatomical landmarks is defined as the current three-marker region.

[0028] Step 402: Based on the three edges of the current three-boundary area, draw lines connecting the midpoint of each edge to the opposite vertex. The intersection of these three lines is determined as the centroid of the current three-boundary area. Specifically, this includes: for each outer enclosing edge, accurately reading the three-dimensional spatial coordinate data corresponding to the two endpoints of the edge, and accurately calculating the three-dimensional coordinates of the unique spatial midpoint of the current edge by averaging the values ​​of the three-dimensional coordinates of the two endpoints. This process is repeated to accurately locate and store the coordinates of all midpoints of the three enclosing edges. For each enclosing edge, accurately matching and determining the corresponding diagonal anatomical boundary. The vertex is the diagonal vertex, which is the remaining anatomical landmark spatial point in the current three-boundary landmark area that does not participate in the formation of the current edge endpoints. The midpoint point obtained from solving each enclosing edge is connected to the corresponding diagonal vertex point by spatial straight lines, generating three independent spatial lines distributed within the current three-boundary landmark area. The three internal spatial lines continue to extend and expand within the current three-boundary landmark area, and all lines eventually converge and coincide at the same fixed spatial point. The unique common intersection and coincidence point of the three spatial lines is retrieved and locked, and this coincidence point is determined as the centroid of the current three-boundary landmark area.

[0029] Step 403: Based on the centroid of the current three-boundary marker area and its three vertices, connect the centroid to the three vertices to obtain three current segmentation lines. Specifically, this includes: retrieving the three-dimensional spatial coordinates of the centroid of the current three-boundary marker area, and simultaneously retrieving the three-dimensional spatial coordinate parameters corresponding to the three anatomical landmark vertices constituting the current three-boundary marker area; using the centroid of the current three-boundary marker area as a unified central starting point, and in an orderly arrangement, performing spatial straight line connection operations with the three vertex points of the current three-boundary marker area one by one, with the three vertex points corresponding to the spatial coordinate points of the three sets of anatomical landmarks in the current CT image; through multiple sets of point-to-point straight line connections, generating three spatial straight lines that are regular in direction, evenly distributed, parallel to each other without intersection, and completely placed inside the current three-boundary marker area. These three sets of spatial straight lines are the current segmentation lines used for region segmentation.

[0030] Step 404: Based on the three current state dividing lines, the current state three-boundary marker area is divided into three current state sub-boundary marker areas, resulting in a current state sub-boundary marker group. Specifically, the three current state dividing lines extend completely from the center point of the current state three-boundary marker area to the three vertices around it, traversing the entire internal space of the current state three-boundary marker area. Relying on the spatial separation effect of the three dividing lines, the overall closed space area of ​​the current state three-boundary marker area is completely divided. After the division, three closed sub-space areas with independent spatial ranges, complete boundary outlines, no overlap, and no gaps are formed. Each independent closed sub-space area can independently correspond to a local lung tissue anatomical area, which can meet the needs of local deformation analysis. This single independent sub-region is the current state sub-boundary marker area. All three current state sub-boundary marker areas obtained from the division are integrated and summarized. All sub-regions are combined to form a complete current state sub-boundary marker group.

[0031] This embodiment achieves corresponding matching of local analysis units in front and behind images, reducing the probability of deformation calculation misalignment caused by differences in regional division standards.

[0032] In a preferred embodiment of the present invention, step 5, calculating the area ratio of the corresponding sub-marker regions in the initial state sub-marker group and the current state sub-marker group to obtain the local deformation coefficient, may include: Step 501: Determine the area of ​​the initial state sub-marker region based on the image range occupied by each initial state sub-marker region in the initial state sub-marker group; determine the area of ​​the current state sub-marker region based on the image range occupied by each current state sub-marker region in the current state sub-marker group. Specifically, this includes: for each independent initial state sub-marker region contained within the initial state sub-marker group, using the complete boundary contour of the corresponding initial state sub-marker region as the limited range, traversing all CT image pixel points covered within the boundary, filtering out valid image pixels belonging to the current initial state sub-marker region that are not background or noise, and completely counting the total number of all valid pixels within a single initial state sub-marker region; relying on the unified image pixel spatial size standard preprocessed by this method, using the actual image plane area corresponding to a single pixel as a fixed conversion base, and determining the area of ​​the single initial state sub-marker region. The total number of effective pixels in the sub-marker area is multiplied by the standard plane area of ​​a single pixel to accurately calculate the actual plane area of ​​the corresponding initial state sub-marker area. The area values ​​of all three initial state sub-marker areas in the initial state sub-marker group are calculated and recorded one by one in the same way to complete the accurate determination of the area of ​​all initial state sub-marker areas. Using the same effective pixel filtering rules, pixel statistics methods and area conversion standards, synchronous calculation is carried out for each independent current state sub-marker area contained in the current state sub-marker group. The pixels inside the boundary of each current state sub-marker area are traversed one by one, effective pixels are filtered and the total number of effective pixels is counted. Combined with the standard plane area of ​​a single pixel, the actual plane area of ​​the corresponding current state sub-marker area is obtained. The accurate determination and value retention of the area of ​​all current state sub-marker areas are completed one by one.

[0033] Step 502: Based on the one-to-one correspondence between the anatomical landmark vertices of the initial state three-boundary marker region and the current state three-boundary marker region, spatially align the initial state sub-marker regions and current state sub-marker regions surrounding the same anatomical landmark vertex, so that the common anatomical landmark vertex coincides and the centroid of the initial state sub-marker region coincides with the centroid of the current state sub-marker region, thus obtaining aligned sub-marker region pairs. Specifically, this includes: establishing a unique correspondence between the landmark vertices of the preceding and following images; the three vertex points of the initial state three-boundary marker region sequentially correspond to the three sets of anatomical landmarks in the preceding image, and the three vertex points of the current state three-boundary marker region sequentially correspond to the three sets of the same anatomical landmarks in the current image; the vertex points corresponding to the same anatomical landmark in the two images are in a one-to-one matching relationship; based on this strict vertex matching relationship, select the initial state sub-marker regions and current state sub-marker regions corresponding to the same anatomical landmark vertex to form the sub-regions to be aligned. Domain combination; For each corresponding sub-region combination, perform precise spatial alignment with two points. The complete alignment process consists of two fixed operations: First, retrieve the three-dimensional coordinates of the anatomical landmark vertices shared by the two sets of sub-landmark areas. Using the spatial position of the common vertex of the current sub-landmark area as a reference, perform a spatial translation transformation on the entire initial sub-landmark area so that the common vertex of the initial sub-landmark area and the common vertex of the current sub-landmark area completely coincide in three-dimensional space. Based on the coincidence of the common vertices, retrieve the three-dimensional coordinates of the centroid of the initial sub-landmark area and the three-dimensional coordinates of the centroid of the current sub-landmark area respectively, and continue to perform a slight spatial translation adaptation on the entire initial sub-landmark area so that the centroid points of the two sets of sub-landmark areas completely coincide in three-dimensional space. After completing the dual-point alignment calibration, the spatial positions of the two sets of sub-regions have achieved maximum adaptation and matching, and finally, the aligned sub-landmark area pair is obtained.

[0034] Step 503: Based on the aligned sub-marker region pairs, determine the intersection area between the initial state sub-marker region and the current state sub-marker region to obtain the intersection area. Calculate the area of ​​the intersection area as the sub-marker region intersection area. Specifically, this includes: using the overall spatial range of the aligned sub-marker region pairs as the search boundary, traversing all image pixel regions covered by the initial state and current state sub-marker regions, pixel by pixel determining whether the current pixel location falls within both the boundary range of the initial state sub-marker region and the boundary range of the current state sub-marker region; and identifying all pixels simultaneously covered by both sets of sub-marker regions that satisfy the dual region affiliation condition. The pixel regions are summarized and integrated. This continuous region with bidirectional overlap and common coverage is the intersection region corresponding to the current sub-boundary region. Using the effective pixel screening criteria in step 501, all invalid noise pixels and background pixels in the intersection region are removed. The total number of all effective image pixels in the intersection region is accurately counted. Then, the total number of effective pixels in the intersection region is multiplied by the actual plane area corresponding to a single pixel to accurately calculate the actual plane area of ​​the intersection region. This area value is defined as the intersection area of ​​the current sub-boundary region and the corresponding sub-boundary region.

[0035] Step 504: Based on the area of ​​the initial state sub-marker region and the intersection area of ​​the sub-marker regions, calculate the ratio of the initial state sub-marker region area to the intersection area of ​​the sub-marker regions. Use this ratio as the local deformation coefficient between the initial state sub-marker region and the current state sub-marker region. Specifically, this includes: for each pair of matched sub-marker regions, retrieve the complete planar area value of the current initial state sub-marker region, and simultaneously retrieve the intersection area value of the corresponding sub-marker regions for that pair. Calculate the corresponding area ratio group by group using the method of dividing the initial state sub-marker region area value by the intersection area value. When scanning the local lung tissue twice... When stretching deformation occurs, the area of ​​the intersection region will be smaller than the area of ​​the initial state sub-landmark region, and the calculated ratio will be greater than one. When local lung tissue undergoes contraction deformation, the area of ​​the intersection region will be close to the area of ​​the initial state sub-landmark region, and the calculated ratio will approach one. This area ratio can accurately quantify the difference in stretching and contraction deformation of a single local lung tissue during two CT scans, and intuitively reflect the degree of non-uniform deformation in different local areas. Finally, the area ratio calculated for each group of sub-regions is uniquely identified as the local deformation coefficient between the initial state sub-landmark region and the corresponding state sub-landmark region.

[0036] In this embodiment, the sub-region matching is completed by statistically analyzing the area and intersection area of ​​the corresponding sub-boundary areas before and after, and by using a dual-point alignment method. The local deformation coefficient is obtained by quantifying the area ratio. This method can better reflect the differences in stretching and contraction deformation of different local lung tissues during two scans, and can adapt to the non-uniform deformation characteristics caused by lung respiratory movements and changes in body position.

[0037] In a preferred embodiment of the present invention, step 6 involves determining the sub-marker region into which the marked nodule location falls within the initial state sub-marker group; mapping the marked nodule location to the current CT image based on the vertex correspondence between the initial state three-marker region and the current state three-marker region to obtain the predicted matching location; and correcting the predicted matching location along the centroid direction of the sub-marker region using the local deformation coefficient corresponding to the sub-marker region to obtain the corrected nodule location, which may include: Step 601: Based on the spatial coordinates of the marked nodule location within the initial three-boundary marker area, determine the initial sub-marker area into which the marked nodule location falls, and obtain the target initial sub-marker area. Specifically, this includes: retrieving the complete three-dimensional spatial coordinate data of the target lung nodule that has been marked in the previous standardized CT image, and clarifying the specific spatial position of the nodule in the overall spatial coordinate system of the initial three-boundary marker area; relying on the complete boundary range and spatial outline of the three initial sub-marker areas that have been segmented, performing regional judgment and verification on the spatial coverage range of the three initial sub-marker areas one by one; by comparing the inclusion relationship between the nodule's three-dimensional coordinates and the boundary spatial range of the three initial sub-marker areas, accurately determining which independent sub-region of the three initial sub-marker areas the current lung nodule's three-dimensional spatial position falls into, excluding the other two initial sub-marker areas that do not cover the nodule position, uniquely locking the single initial sub-marker area that can completely contain the target lung nodule coordinate position, and defining the precisely matched and located initial sub-marker area as the target initial sub-marker area.

[0038] Step 602: Based on the one-to-one correspondence between the three vertices of the initial three-boundary marker area and the three vertices of the current three-boundary marker area, the relative position of the marked nodule within the initial three-boundary marker area is mapped to the current three-boundary marker area to obtain the predicted matching position. Specifically, this includes: relying on the one-to-one correspondence between the three anatomical landmark vertices of the initial three-boundary marker area and the three anatomical landmark vertices of the current three-boundary marker area, the corresponding vertices of the two sets of three-boundary marker areas are the spatial positions of the same anatomical structure in the two images; performing the nodule relative position parameter solution operation, constructing a local spatial reference system with the three vertices of the initial three-boundary marker area, calculating the spatial distance parameters of the three-dimensional coordinates of the marked lung nodule relative to each vertex of the initial three-boundary marker area, and combining the overall spatial span of the three-boundary marker area to calculate the... The spatial proportion parameters and spatial weight ratio parameters of the lung nodule location relative to the three vertices within the initial three-boundary marker area are used to fully record the refined relative position parameters of the nodule within the initial three-boundary marker area. This set of parameters is not affected by overall regional deformation and spatial displacement and has a unique correspondence. The fixed set of relative position ratio parameters is fully mapped and matched to the three corresponding vertices of the current three-boundary marker area. Based on the overall spatial span and boundary range of the current three-boundary marker area itself, combined with the aforementioned relative position ratio parameters, the new three-dimensional coordinate points corresponding to the ratio parameters in the spatial system of the current three-boundary marker area are calculated and solved dimension by dimension. The new spatial points obtained by ratio mapping and cross-frame conversion are the predicted matching positions of the lung nodule in the current CT image.

[0039] Step 603: Based on the centroid of the target initial state sub-landmark area and the position of the marked nodule, the direction from the centroid of the target initial state sub-landmark area to the position of the marked nodule is determined as the offset direction. Specifically, this includes: retrieving the complete three-dimensional spatial coordinate parameters of the centroid of the target initial state sub-landmark area, and simultaneously retrieving the precise three-dimensional spatial coordinate parameters of the marked target lung nodule in the previous CT image; using the three-dimensional point of the centroid of the target initial state sub-landmark area as the spatial starting base point and the three-dimensional point of the marked lung nodule as the spatial ending base point, establishing a spatial connection between the two points in the three-dimensional image space, and completely determining the three-dimensional vector direction of the spatial connection; this vector direction can accurately characterize the spatial offset trend of the target lung nodule relative to its local centroid, and uniformly determining the fixed three-dimensional vector direction from the centroid to the nodule marking point as the offset direction of the nodule relative to the centroid.

[0040] Step 604: Based on the local deformation coefficient and offset direction corresponding to the target initial state sub-marker region, the distance between the predicted matching position and the centroid of the target initial state sub-marker region is scaled and corrected along the offset direction using the local deformation coefficient to obtain the corrected nodule position. Specifically, this includes: retrieving the unique local deformation coefficient value corresponding to the target initial state sub-marker region, and combining it with the fixed offset direction determined in step 603 to complete the refined deformation correction processing of the predicted matching position; obtaining the three-dimensional coordinates of the predicted matching position and the three-dimensional coordinates of the centroid of the target initial state sub-marker region, calculating the straight-line distance between the two points in three-dimensional space, and obtaining the uncorrected nodule position. The original spatial distance value is obtained. This original spatial distance value is multiplied by the local deformation coefficient value corresponding to the target initial state sub-marker area. By multiplying the distance by the deformation coefficient, the deformation characteristics of local lung tissue stretching or contracting are adapted to obtain the new corrected distance value after local deformation compensation. After the distance correction is completed, the centroid of the target initial state sub-marker area is used as a fixed starting point. The new corrected distance value is extended along the previously determined nodule offset direction. The new three-dimensional spatial point that is finally locked is the corrected nodule position after eliminating local deformation deviation.

[0041] This embodiment determines the target sub-marker area to which the nodule belongs, completes the initial nodule position mapping by combining the relative position ratio of the markers, and then completes the position correction by relying on the local deformation coefficient and the nodule offset direction; it can fine-tune the predicted nodule position by combining the actual local deformation, which can effectively alleviate the nodule position offset problem caused by local lung deformation.

[0042] In a preferred embodiment of the present invention, step 7, constructing a spherical search area centered on the corrected nodule location and with the labeled nodule diameter as the radius, and performing nodule detection on the current CT image within the spherical search area to complete the automatic detection and comparison of lung nodules, may include: Step 701: Based on the corrected nodule location and the labeled nodule diameter, a spherical search area is constructed with the corrected nodule location as the center and the labeled nodule diameter as the radius. Specifically, this includes: retrieving the complete three-dimensional spatial coordinate parameters of the corrected nodule location, and simultaneously retrieving the measured diameter value of the target lung nodule manually labeled and retained from the previous CT image; using the three-dimensional spatial point of the corrected nodule location as the reference point of the spherical geometric center, and the measured diameter value of the labeled lung nodule as the spherical radiation radius, the entire area is filled and constructed according to the standard spherical geometric configuration within the three-dimensional spatial system of the preprocessed standardized CT image, generating a three-dimensional spherical closed space region with complete boundaries, fixed radius, and precise center; this spherical region is constructed strictly based on the actual size of the nodule and the precise point after correction, and its range closely matches the actual size and actual position of the nodule after deformation, accurately covering the possible displacement area of ​​the nodule. This specially constructed spherical closed space region is the spherical search area.

[0043] Step 702: Based on the spherical search area, nodule detection is performed on the image area covered by the spherical search area in the current CT image to obtain a candidate nodule set. Specifically, this includes: for the current CT image that has been preprocessed and has unified spatial scale and grayscale standards, based on the three-dimensional spatial range of the spherical search area, local three-dimensional image volume data completely covered by the spherical search area is extracted, and all irrelevant image areas outside the spherical search area are removed. Only the local image volume data within the limited range is retained for subsequent detection operations, effectively reducing the detection range and avoiding redundant calculations and interference noise in full-domain detection; a fully trained three-dimensional convolutional neural network nodule detection model is used to perform nodule recognition. The model is fully built using a three-dimensional U-Net architecture. The overall network structure is composed of three-dimensional downsampling layers composed of multiple layers of three-dimensional convolution and three-dimensional upsampling layers composed of multiple layers of three-dimensional deconvolution stacked in an orderly manner. A one-to-one skip connection structure is set between the output feature map of each level of the downsampling layer and the input feature map of the upsampling layer.The three-dimensional downsampling layer progressively extracts features and compresses scale from the input CT local image volume data, capturing shallow texture details and deep semantic features layer by layer, and extracting morphological, grayscale, and edge three-dimensional feature information of lung nodules at multiple scales. The three-dimensional upsampling layer progressively restores scale and reconstructs features from the compressed deep feature map, gradually restoring the original spatial resolution of the feature map and ensuring the accuracy of feature localization. The same-level skip connection structure fuses shallow fine detail features and deep abstract semantic features pixel by pixel, making up for the details lost during downsampling and significantly improving the recognition accuracy of small nodules and nodules with blurred edges. During the model training phase, a large number of chest CT thin-slice image samples acquired from multiple centers and multiple devices are collected. All samples are accurately manually annotated by professional radiologists to determine the location, outline, and size of lung nodules, constructing a standardized lung nodule annotation sample dataset. All training sample images are uniformly subjected to the full set of preprocessing operations corresponding to this method, such as layer thickness resampling and grayscale normalization, to ensure that the processing standards of training data and actual detection data are consistent. The inter-scale and grayscale systems are completely consistent. The preprocessed 3D image sample data and the corresponding nodule real-label data are synchronously input into the 3D U-Net model. The error difference between the model's predicted nodule region and the artificially labeled region is calculated pixel by pixel using the cross-entropy loss function. The error backpropagation mechanism is used to correct and iteratively update all convolution parameters and weight parameters inside the model layer by layer. The training process is continuously iterated to reduce the model's prediction error until the overall model loss value tends to be stable and there is no obvious downward trend after multiple iterations. The model training is then determined to be converged, and the optimal model weight parameters are fixed and saved to complete the model training. In the actual detection and inference stage, the local 3D image volume data of the spherical search area is input into the converged and fixed optimal 3D U-Net model. The model outputs the 3D regions and corresponding point information of all candidate nodules in the current local area that meet the 3D image characteristics of lung nodules through a complete feature extraction, feature fusion, and region inference process. All effectively identified candidate nodule data are summarized and integrated to form a candidate nodule set.

[0044] Step 703: Based on the morphological features of each candidate nodule in the candidate nodule set and the morphological features of the labeled nodules, calculate the similarity between each candidate nodule and the labeled nodules, and select the candidate nodule with the highest similarity as the matching nodule to complete the automatic detection and comparison of lung nodules. Specifically, this includes: extracting multiple core morphological features from the three-dimensional image region of each candidate nodule, including the overall size of the nodule, the regularity of the three-dimensional morphological contour, the uniformity of the internal image grayscale distribution, the clarity of edge texture and edge spiculation features, the spatial proportion of the nodule, etc., forming independent multi-dimensional feature vector data for each candidate nodule; simultaneously retrieving the image data of the labeled target lung nodules from the previous CT images, and extracting the features using completely consistent feature extraction rules and dimensions. The system obtains a complete set of multidimensional feature vector data corresponding to the previously labeled nodules, and constructs a feature data set that can be accurately compared before and after. Using a unified feature similarity quantification calculation rule, the system compares the corresponding feature parameters of the candidate nodules with those of the labeled nodules dimension by dimension, calculates the overall matching similarity value of the two sets of multidimensional feature vectors, and quantifies the degree of morphological fit between each candidate nodule and the original follow-up nodule. After completing the similarity quantification calculation of all candidate nodules, the system sorts and filters all similarity values, extracts the single candidate nodule with the highest similarity value among all candidate nodules, and determines the candidate nodule with the highest matching degree as the matching target nodule in the current image corresponding to the lung nodule in the previous follow-up. Finally, the system accurately completes the automatic follow-up detection and comparison matching of lung nodules between two chest CT images of the same subject.

[0045] In this embodiment, a spherical search area is constructed based on the corrected nodule location and the original nodule size, which can reasonably reduce the image detection range and achieve localized targeted detection. This method can reduce the computational resource consumption caused by global detection to a certain extent and reduce unnecessary false positive interference.

[0046] like Figure 2 As shown, embodiments of the present invention also provide an automatic detection and comparison system for lung nodules in chest CT images, comprising: The acquisition module is used to acquire current and previous CT images of the same subject. The previous CT images include the marked nodule locations and marked nodule diameters. The extraction module is used to extract the tracheal carina, the bifurcation of the right upper lobe bronchus, and the upper margin of the left pulmonary artery arch from the lung parenchyma region of the previous CT image as three anatomical landmarks. The construction module is used to construct the initial three-boundary region based on the coordinates of the three anatomical landmarks in the previous CT image. The initial three-boundary region is divided by connecting the centroid of the initial three-boundary region to the three sides of the initial three-boundary region to obtain the initial sub-boundary group. The segmentation module is used to construct the current three-boundary region based on the coordinates of the three anatomical landmarks in the current CT image. The current three-boundary region is segmented by connecting the centroid of the current three-boundary region to the three sides of the current three-boundary region to obtain the current sub-boundary group. The calculation module is used to calculate the area ratio of the corresponding sub-marker regions in the initial state sub-marker group and the current state sub-marker group, and obtain the local deformation coefficient. The correction module is used to determine the sub-marker region in which the marked nodule location falls within the initial state sub-marker group; based on the vertex correspondence between the initial state three-marker region and the current state three-marker region, the marked nodule location is mapped to the current CT image to obtain the predicted matching location; along the centroid direction of the sub-marker region, the predicted matching location is corrected using the local deformation coefficient corresponding to the sub-marker region to obtain the corrected nodule location; The detection module is used to construct a spherical search area with the corrected nodule position as the center and the labeled nodule diameter as the radius. Within the spherical search area, nodule detection is performed on the current CT image to complete the automatic detection and comparison of lung nodules.

[0047] It should be noted that this system is a system corresponding to the above method. All implementation methods in the above method embodiments are applicable to this embodiment and can achieve the same technical effect.

[0048] Embodiments of the present invention also provide a computing device, including: a processor and a memory storing a computer program, wherein the computer program, when executed by the processor, performs the method described above. All implementations in the above method embodiments are applicable to this embodiment and can achieve the same technical effects.

[0049] Embodiments of the present invention also provide a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the method described above. All implementations in the above method embodiments are applicable to this embodiment and can achieve the same technical effects.

[0050] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. An automatic detection and comparison method for lung nodules in chest CT images, characterized in that, The method includes: Step 1: Obtain the current CT image and the previous CT image of the same subject. The previous CT image includes the marked location and diameter of the nodules. Step 2: Extract the tracheal carina, the bifurcation of the right upper lobe bronchus, and the upper margin of the left pulmonary artery arch from the lung parenchyma region of the previous CT image as three anatomical landmarks. Step 3: Construct the initial three-boundary region based on the coordinates of the three anatomical landmarks in the previous CT image. Divide the initial three-boundary region by connecting the centroid of the initial three-boundary region to the three sides of the initial three-boundary region to obtain the initial sub-boundary group. Step 4: Construct the current three-boundary region based on the coordinates of the three anatomical landmarks in the current CT image. Divide the current three-boundary region by connecting the centroid of the current three-boundary region to the three sides of the current three-boundary region to obtain the current sub-boundary group. Step 5: Calculate the area ratio of the corresponding sub-marker regions in the initial state sub-marker group and the current state sub-marker group to obtain the local deformation coefficient; Step 6: Determine the sub-marker region into which the marked nodule location falls within the initial state sub-marker group; based on the vertex correspondence between the initial state three-marker region and the current state three-marker region, map the marked nodule location onto the current CT image to obtain the predicted matching location; along the centroid direction of the sub-marker region, correct the predicted matching location using the local deformation coefficient corresponding to the sub-marker region to obtain the corrected nodule location, including: determining the initial state sub-marker region into which the marked nodule location falls based on the spatial coordinates of the marked nodule location within the initial state three-marker region to obtain the target initial state sub-marker region; based on the three vertices of the initial state three-marker region... The one-to-one correspondence between the point and the three vertices of the current state three-boundary marker region maps the relative position of the labeled nodule in the initial state three-boundary marker region to the current state three-boundary marker region, thus obtaining the predicted matching position. Based on the centroid of the target initial state sub-boundary marker region and the labeled nodule position, the direction from the centroid of the target initial state sub-boundary marker region to the labeled nodule position is determined as the offset direction. Based on the local deformation coefficient corresponding to the target initial state sub-boundary marker region and the offset direction, the distance between the predicted matching position and the centroid of the target initial state sub-boundary marker region is scaled and corrected along the offset direction using the local deformation coefficient, thus obtaining the corrected nodule position. Step 7: Construct a spherical search area with the corrected nodule location as the center and the marked nodule diameter as the radius. Perform nodule detection on the current CT image within the spherical search area to complete the automatic detection and comparison of lung nodules.

2. The automatic detection and comparison method for lung nodules in chest CT images according to claim 1, characterized in that, Three anatomical landmarks were extracted from the lung parenchyma region of the previous CT image: the carina, the bifurcation of the right upper lobe bronchus, and the upper margin of the left pulmonary artery arch. Based on the density difference between lung tissue and surrounding tissue in the previous CT image, the lung parenchyma region is obtained by density threshold segmentation; based on the density difference between the tracheal wall and the air in the tracheal lumen within the lung parenchyma region, the tracheal tree structure is extracted. Based on the tracheal tree structure, the tracheal lumen is sectioned layer by layer along the head-to-foot direction to obtain a sequence of tracheal lumen sections. Based on the change in the number of tracheal lumen sections at each level in the tracheal lumen section sequence, the first level where the number of tracheal lumen sections changes from one to two is identified as the tracheal carina. Based on the tracheal lumen section on the right side of the plane where the carina is located, the starting section of the right main bronchus is determined; based on the tracheal lumen section on the left side of the plane where the carina is located, the starting section of the left main bronchus is determined; based on the starting section of the right main bronchus, the outline of the right main bronchus section is traced layer by layer along the head-to-foot direction, and the plane where the first branch opening appears on the outer wall of the right main bronchus section is determined as the bifurcation point of the right upper lobe bronchus; Based on the vascular morphology at the left hilum level in the lung parenchyma, the course of the left pulmonary artery was extracted, and the upper margin of the arched vessel above the left main bronchus was determined as the upper margin of the left pulmonary artery arch. Based on the carina position, the bifurcation position of the right upper lobe bronchus, and the upper margin of the left pulmonary artery arch, three anatomical landmarks were obtained.

3. The automatic detection and comparison method for lung nodules in chest CT images according to claim 2, characterized in that, Based on the coordinates of the three anatomical landmarks in the previous CT image, an initial three-landmark region is constructed. The initial three-landmark region is then divided by connecting the centroid of the initial three-landmark region to its three sides, resulting in an initial sub-landmark group, including: Based on the three-dimensional coordinates of the three anatomical landmarks in the previous CT image, connect the three anatomical landmarks pairwise with straight lines to obtain the initial three-landmark region enclosed by three sides; based on the three sides of the initial three-landmark region, draw a line connecting the midpoint of each side to the opposite vertex, and determine the centroid of the initial three-landmark region by the intersection of the three lines. Connect the centroid of the initial state three-boundary marker region with the three vertices of the initial state three-boundary marker region to obtain three initial state dividing lines; based on the three initial state dividing lines, divide the initial state three-boundary marker region into three initial state sub-boundary marker regions to obtain the initial state sub-boundary marker group.

4. The automatic detection and comparison method for lung nodules in chest CT images according to claim 3, characterized in that, Based on the coordinates of the three anatomical landmarks within the current CT image, a current three-landmark region is constructed. The current three-landmark region is then divided by connecting the centroid of the current three-landmark region to its three sides, resulting in a current sub-landmark group, including: Based on the three-dimensional coordinates of the three anatomical landmarks in the current CT image, connect the three anatomical landmarks pairwise with straight lines to obtain the current three-landmark region enclosed by three sides; Based on the three sides of the current three-boundary region, draw a line connecting the midpoint of each side to the opposite vertex, and determine the intersection of the three lines as the centroid of the current three-boundary region; Based on the centroid of the current state three-boundary marker region and the three vertices of the current state three-boundary marker region, connect the centroid to the three vertices respectively to obtain three current state dividing lines; Based on the three current state dividing lines, the current state three-boundary marker region is divided into three current state sub-boundary marker regions, resulting in the current state sub-boundary marker group.

5. The automatic detection and comparison method for lung nodules in chest CT images according to claim 4, characterized in that, Calculate the area ratio of the corresponding sub-marker regions in the initial state sub-marker set and the current state sub-marker set to obtain the local deformation coefficient, including: The area of ​​the initial state sub-marker region is determined based on the image range occupied by each initial state sub-marker region in the initial state sub-marker group; the area of ​​the current state sub-marker region is determined based on the image range occupied by each current state sub-marker region in the current state sub-marker group. Based on the one-to-one correspondence between the initial state three boundary marker regions and the current state three boundary marker regions, the initial state sub-boundary marker regions and the current state sub-boundary marker regions surrounding the same anatomical boundary marker vertex are spatially aligned so that the common anatomical boundary marker vertex coincides and the centroid of the initial state sub-boundary marker region coincides with the centroid of the current state sub-boundary marker region, thus obtaining the aligned sub-boundary marker region pair. Based on the alignment of the sub-marker regions, determine the intersection region between the initial state sub-marker region and the current state sub-marker region to obtain the intersection region. Calculate the area of ​​the intersection region as the sub-marker region intersection area. Based on the area of ​​the initial state sub-boundary region and the area of ​​the intersection of the sub-boundary regions, calculate the ratio of the area of ​​the initial state sub-boundary region to the area of ​​the intersection of the sub-boundary regions, and use this ratio as the local deformation coefficient between the initial state sub-boundary region and the current state sub-boundary region.

6. The automatic detection and comparison method for lung nodules in chest CT images according to claim 5, characterized in that, A spherical search area is constructed with the corrected nodule location as the center and the labeled nodule diameter as the radius. Nodule detection is performed on the current CT image within this spherical search area, completing the automatic detection and comparison of lung nodules, including: Based on the corrected nodule location and the labeled nodule diameter, a spherical search area is constructed with the corrected nodule location as the center and the labeled nodule diameter as the radius. Based on the spherical search area, nodules are detected in the image area covered by the spherical search area in the current CT image to obtain a set of candidate nodules; Based on the morphological features of each candidate nodule in the candidate nodule set and the morphological features of the labeled nodules, the similarity between each candidate nodule and the labeled nodules is calculated, and the candidate nodule with the highest similarity is selected as the matching nodule to complete the automatic detection and comparison of lung nodules.

7. An automatic detection and comparison system for lung nodules on chest CT images, wherein the system implements the method as described in any one of claims 1 to 6, characterized in that, include: The acquisition module is used to acquire current and previous CT images of the same subject. The previous CT images include the marked nodule locations and marked nodule diameters. The extraction module is used to extract the tracheal carina, the bifurcation of the right upper lobe bronchus, and the upper margin of the left pulmonary artery arch from the lung parenchyma region of the previous CT image as three anatomical landmarks. The construction module is used to construct the initial three-boundary region based on the coordinates of the three anatomical landmarks in the previous CT image. The initial three-boundary region is divided by connecting the centroid of the initial three-boundary region to the three sides of the initial three-boundary region to obtain the initial sub-boundary group. The segmentation module is used to construct the current three-boundary region based on the coordinates of the three anatomical landmarks in the current CT image. The current three-boundary region is segmented by connecting the centroid of the current three-boundary region to the three sides of the current three-boundary region to obtain the current sub-boundary group. The calculation module is used to calculate the area ratio of the corresponding sub-marker regions in the initial state sub-marker group and the current state sub-marker group, and obtain the local deformation coefficient. The correction module is used to determine the sub-marker region in which the marked nodule location falls within the initial state sub-marker group; based on the vertex correspondence between the initial state three-marker region and the current state three-marker region, the marked nodule location is mapped to the current CT image to obtain the predicted matching location; along the centroid direction of the sub-marker region, the predicted matching location is corrected using the local deformation coefficient corresponding to the sub-marker region to obtain the corrected nodule location; The detection module is used to construct a spherical search area with the corrected nodule position as the center and the labeled nodule diameter as the radius. Within the spherical search area, nodule detection is performed on the current CT image to complete the automatic detection and comparison of lung nodules.

8. A computing device, characterized in that, include: One or more processors; A storage device for storing one or more programs that, when executed by one or more processors, cause the one or more processors to implement the method as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a program that, when executed by a processor, implements the method as described in any one of claims 1 to 6.

Images