Source applicator reconstruction method

Abstract

The invention belongs to the technical field of metal source applicator feature extraction under CT images, and discloses a source applicator reconstruction method. CT data of all pipelines of the source applicator are obtained through thin-layer CT scanning, curvilinear equations of all the pipelines are fitted, and a thin-layer CT-based source applicator template library is established; an improved sparse cascade network model is established, so that the source applicators can be quickly and accurately classified accurately, and a basis is provided for further matching corresponding source applicator pipelines in a source applicator template library subsequently; on the basis of the thin-layer CT, the CT-CT image registration form between the CT data in the source applicator template library based on the thin-layer CT and the CT data of the clinical source applicator is more efficient and faster; a source applicator fitting curve established by thin-layer CT data is used as a skeleton, a fitting curve of a clinical source applicator CT data extraction center point set is optimized by means of segmented fitting, physical constraint and the like, and then accurate reconstruction of the source applicator is completed.

Description

Technical Field

[0001] The present invention relates to the technical field of metal applicator feature extraction under CT images, and in particular to a method for reconstructing an applicator. Background Art

[0002] Brachytherapy is a crucial treatment modality in radiotherapy, particularly for gynecological cancers. Different applicators are used in brachytherapy depending on the condition and treatment objectives. During the brachytherapy planning phase of modern radiotherapy, dosimetrists are often required to manually identify the applicator center point for each slice on CT images, a process known as applicator reconstruction. However, applicator reconstruction is not only time-consuming and labor-intensive, but also subject to human error, resulting in various deviations in the reconstructed applicator, such as center point position deviation due to metal artifacts and applicator tip position deviation due to slice thickness. Furthermore, the Fletcher applicator, commonly used in gynecological cancer treatment, is made of metal and appears as a bright feature on CT images, resulting in significant metal artifact noise. Furthermore, its clinical assembly requires the assembly of an ovulation device and a uterine os limiter. These components also contain metal rings and bolts, which often overlap with the applicator's bright features on CT images. Accurately identifying the applicator channel center point requires considerable experience and meticulousness on the part of the dosimetrist, posing a potential risk to treatment safety.

[0003] The related literature (Zhang D, Yang Z, Jiang S, Zhou Z, Meng M, Wang W. Automaticsegmentation and applicator reconstruction for CT-based brachytherapy ofcervical cancer using 3D convolutional neural networks. J Appl Clin Med Phys. 2020 Oct;21(10):158-169. doi: 10.1002 / acm2.13024. Epub 2020 Sep 29. PMID:32991783; PMCID: PMC7592978) introduced an automatic organ and applicator reconstruction method based on three-dimensional convolutional neural networks. In this article, gynecological tumor cases treated with the Fletcher applicator, a commonly used clinical device, were scanned with CT scans for each case. The CT data was enhanced before model training. An improved three-dimensional convolutional neural network was proposed to segment the input CT data into the corresponding contour organs and applicators. The automatic reconstruction accuracy of the applicator was high, with the Dess coefficient exceeding 88% and the Jade coefficient exceeding 80%. However, the article did not mention the assembly of the applicator used, that is, it did not explain the assembly of the uterine cavity limit buckle in the uterine cavity tube. The applicator reconstruction described in this article uses skeletonization and nonlinear fitting. This method is relatively simple, but in actual use, it is difficult to distinguish metal bolts, etc., and there is a certain reconstruction deviation.

[0004] Chinese patent CN116152437A, "Applicator reconstruction method, device, electronic device, and computer-readable storage medium," describes a method that first uses methods such as rotating and cutting CT images and applies a U-net network to obtain and generate a three-dimensional mask. This method does not mention the assembly of the applicator used, that is, it does not explain the assembly of the uterine cavity limit buckle in the uterine canal. The method also does not explain the identification and classification of the left and right oogonium during the reconstruction process, which would lead to potential deviations in the applicator reconstruction. This method uses a three-dimensional mask, but does not explicitly mention whether the mask is theoretical or based on the actual applicator structure. Summary of the Invention

[0005] The purpose of the present invention is to provide a method for reconstructing an applicator, so as to achieve rapid and accurate reconstruction of an applicator used in clinical treatment.

[0006] The technical solution of the present invention is as follows: a source applicator reconstruction method, which obtains thin-slice CT data of each pipeline of the source applicator through thin-slice CT scanning, fits the curve equation of each pipeline, and establishes a source applicator template library based on thin-slice CT;

[0007] An improved sparse cascade network model was established and trained using CT images of applicators under conventional scanning conditions. This model was used to classify clinical applicators into single or multiple tubes. In the case of multiple tubes, the tubes were split based on the classification results to obtain multiple single tubes. Single tubes were matched to applicators in the applicator template library based on thin-slice CT.

[0008] The CT images in the thin-slice CT-based applicator template library and the CT images of the clinical applicator are registered using a multiple registration method based on an improved physical structure-driven optimization; using the applicator fitting curve established in the thin-slice CT-based applicator template library as the skeleton, the fitting curve of the central point set extracted from the CT image of the clinical applicator is optimized using segmented fitting and physical constraints to complete the accurate reconstruction of the applicator.

[0009] The process of acquiring the template library of the applicator based on thin-slice CT is as follows: obtaining a thin-slice CT image of only the canal portion of the applicator; the canal portion includes the left oval canal, the right oval canal, the 15-degree uterine canal, the 30-degree uterine canal, the 45-degree uterine canal, and the 60-degree uterine canal;

[0010] Based on the above thin-slice CT images, the center point set extraction and curve fitting of each type of applicator are performed based on hybrid geometric features; the above thin-slice CT images and fitting curve results are stored as an applicator template library based on thin-slice CT.

[0011] The specific implementation process of the source center point set extraction and curve fitting based on hybrid geometric features is as follows:

[0012] The thin-slice CT images were preprocessed using Gaussian filtering denoising method;

[0013] The dynamic threshold method is used to extract the three-dimensional center point of the applicator pipeline in the thin-slice CT image and obtain the center point set;

[0014] Sorting the point cloud along the applicator path, identifying the endpoints of the applicator center point set, and enhancing the number of point sets at the applicator endpoints;

[0015] Performing curve fitting based on a spline function on the above-mentioned central point set to obtain a fitting curve function;

[0016] Based on the above fitting curve function, curvature analysis and segmentation are performed, and the segmentation equation is as follows;

[0017] (1)

[0018] is the partition equation; is the equation of the fitting curve of the three-dimensional center point of the applicator pipeline; is the arc length, and They are The first and second derivatives of ;

[0019] Based on the curve fitting of the geometric morphology of the applicator, the three-segment smooth connection of the curve is enhanced, and the geometric morphological features of straight line segment, curved segment and straight line segment are obtained in sequence.

[0020] The thin-slice CT image is CT image data with a slice thickness of 0.6 mm.

[0021] The improved sparse cascade network model includes a dynamic sparse activation module, a light branch module for processing simple areas, a deep branch module for processing complex areas, and a cascade decision maker module; the data processed by the deep branch module and the light branch module are feature spliced, and the data after feature splicing passes through a cascade decision maker module, and the high-confidence results are directly output, and the low-confidence results are output after passing through a transformer module; the CT image of the clinical applicator is input, and the applicator types include: single-tube applicator: 15-degree uterine canal, 30-degree uterine canal, 45-degree uterine canal, and 60-degree uterine canal; two-tube applicator: an applicator used in combination with the left oval canal and the right oval canal; three-tube applicator: an applicator used in combination with the left oval canal and the right oval canal and various angles of the uterine canal.

[0022] The input data of the improved sparse cascade network model first passes through the dynamic sparse activation module; the dynamic sparse activation module generates a mask by a gated convolutional layer with a dynamic threshold set to 30% to identify key areas;

[0023] In the above process, the mask generated by the gated convolution layer is regarded as a weight map, which is used together with the dynamic threshold to process the input data. After processing the weight map, two data of the same size are output; the weight values ​​are arranged from large to small; the values ​​of the points before 30% of the weight value are retained, and the values ​​of the other points are set to 0, and this area is defined as a complex area; the values ​​of the points after 30% of the weight value are retained, and the values ​​of the other points are set to 0, and this area is defined as a simple area;

[0024] The simple area output by the dynamic sparse activation module uses the lightweight branch module to extract features; the complex area output by the dynamic sparse activation module uses the deep branch module to extract features;

[0025] The features obtained from the lightweight branch module and the features obtained from the deep branch are concatenated and then passed through a cascade decision maker module; the first level of the cascade decision maker module is global average pooling, connected to a fully connected layer, and the second level is a Transformer encoder block; if the confidence level obtained after the first level is ≥0.85, the classification result is given directly without passing through the second level; when the confidence level is <0.85, the final classification is completed through the second-level Transformer encoder block, and the classification result is output.

[0026] The registration specifically includes: after preprocessing the CT image of the clinical applicator, matching the corresponding pipeline data in the applicator template library based on thin-slice CT according to the name of each pipeline of the applicator, and using a multiple registration method based on improved physical structure driven optimization to complete the three-dimensional spatial registration of the CT image of the clinical applicator with the CT image of the corresponding applicator in the applicator template library based on thin-slice CT, and establish the corresponding relationship between the two applicators in the three-dimensional space;

[0027] The reconstruction of the clinical applicator is specifically as follows: calling the center point set fitting curve function of the applicator template library based on thin-layer CT, correcting the center point set on the clinical applicator CT image based on the above-mentioned alignment results, and creating the applicator outline on the clinical applicator CT image according to the physical size of the applicator to complete the reconstruction of the clinical applicator.

[0028] The specific implementation process of the multiple registration method based on improved physical structure driven optimization is as follows:

[0029] Initial image registration: The bone-based registration method is used to complete the approximate registration of the CT image of the corresponding applicator in the thin-slice CT-based applicator template library with the clinical applicator CT image;

[0030] The fitting curve obtained from the applicator template library based on thin-slice CT is used as the "shape skeleton" to guide the point set formed by the clinical applicator CT image to fit within the target range;

[0031] Extraction and fitting of the center point set of the clinical applicator to strengthen the three-segment structural features; the three-segment structural features are straight segment, curved segment, and straight segment in sequence;

[0032] Weighted registration: The weight of the curved segment is enhanced, and the registration of the curved segment is strengthened in the multi-segment registration, and finally the three-dimensional spatial registration of the clinical applicator and the corresponding applicator in the applicator template library based on thin-slice CT is achieved. The objective function of the relevant registration is given by formula (2);

[0033] The physical dimensions of the pipeline are reconstructed through constrained fitting of the center point set of the clinical applicator.

[0034] (2)

[0035] in, is the constraint weight of the line segment, It represents the distance between the point corresponding to the straight line segment extracted from the clinical applicator CT image after displacement and the straight line segment in the CT image of the applicator template library based on thin-slice CT. is the corresponding point in the straight line segment after displacement, is a straight line segment in thin-slice CT, is the constraint weight of the curved segment, is the corresponding point in the curved line segment after displacement, It is the corresponding point set in the curved segment of the CT image of the applicator template library based on thin-slice CT; It represents the sum of the distances from the point corresponding to the straight line segment extracted from the clinical applicator CT image after displacement to the straight line segment in the CT image of the applicator template library based on thin-slice CT. represents the sum of the minimum values ​​of the distances between the points corresponding to the curved segments extracted from the clinical applicator CT image after displacement and the curved segments in the CT images of the applicator template library based on thin-slice CT;

[0036] Formula (2) is the objective function of weighted registration, which aims to optimize the rigid transformation parameters so that the center point set extracted from the clinical applicator CT image is aligned with the segmented geometric features of the fitting curve corresponding to the applicator template library based on thin-slice CT;

[0037] (3)

[0038] Formula (3) is the energy function of the smoothing spline, which is used for piecewise constrained fitting. Under the condition of known piecewise structure and transition point constraints, smooth curve fitting is performed;

[0039] in: It is a fitting curve function in the template library of the applicator based on thin-slice CT. The fitting function of the center point set of the clinical applicator CT image, is a parameter; is the weight, which is greater at the bend; is the smoothing coefficient; The center point of the clinical applicator CT image is located in the straight line segment. Indicates that the center points in the applicator template library based on thin-slice CT are concentrated at the intersection of the straight line segment and the curve segment, The function representing the center point set of the clinical applicator CT image at the intersection of the straight line-curve and the curve-straight line, is the constraint strength coefficient;

[0040] Apply position constraints at the transition points:

[0041]

[0042] in, and It is the position of the transition point of the fitting curve of the applicator template library based on thin-slice CT.

[0043] Compared with the existing technology, the technical solution proposed in the present invention obtains high-resolution three-dimensional data of the source through thin-layer CT (0.6mm) scanning, which is more efficient, quicker and more accurate in subsequent image registration, curve fitting optimization and other aspects. In actual clinical use, the source applicator generally requires multiple channels to be used in combination according to treatment needs. The present invention selects the corresponding source channel CT data from the source applicator template library based on thin-layer CT according to the type and model of the clinically used source applicator. The two are subjected to multi-level registration, and the fitting curve obtained from the source applicator in the source applicator template library based on thin-layer CT is used as the skeleton and combined with the geometric characteristics of the source applicator. In the registration and fitting process, the fitting optimization of the center point set of the clinical source applicator is achieved through conditional constraints, thereby realizing the accurate reconstruction of the clinically used source applicator. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 It is the overall flow chart of the technical solution of the present invention;

[0045] Figure 2 A flowchart for establishing a template library of applicators based on thin-slice CT;

[0046] Figure 3 To improve the flow chart of sparse cascade network model;

[0047] Figure 4 To improve the structure diagram of sparse cascade network model;

[0048] Figure 5 This is the structural diagram of the dynamic sparse activation module;

[0049] Figure 6 This is a diagram of the lightweight branch module structure;

[0050] Figure 7 This is the structure diagram of the deep branch module;

[0051] Figure 8 It is the structural diagram of cascade decision module;

[0052] Figure 9 This is the registration flow chart. DETAILED DESCRIPTION

[0053] like Figure 2 As shown in the figure, a template library of applicators based on thin-slice CT is established:

[0054] Obtain thin-slice CT data of the applicator without accessories such as metal bolts, that is, only the canal portion (left oval canal, right oval canal, 15-degree uterine canal, 30-degree uterine canal, 45-degree uterine canal, and 60-degree uterine canal). The present invention uses CT image data with a slice thickness of 0.6 mm.

[0055] Based on the above thin-slice CT images, the center point set of each applicator model was extracted and fitted based on hybrid geometric features.

[0056] The thin-slice CT images and fitting curve results are saved as XML format files and stored as a template library of applicators based on thin-slice CT;

[0057] like Figure 3 As shown, an improved sparse cascade network model is established. The applicators used in clinical treatment are assembled according to different treatment purposes. The present invention uses an improved sparse cascade network (Sparse-Cascade Net) to identify and classify the applicators used in clinical treatment;

[0058] CT image preprocessing of clinical single-channel / multi-channel applicators: After identification and classification, the applicator retains its original geometric characteristics and uses a three-dimensional feature extraction method to process the CT image, including removing noise and foreign matter. However, when the applicator is detected to be multi-channel, the applicator needs to be split to ultimately form multiple single-tube applicator CT image data.

[0059] Call the applicator CT image in the applicator template library based on thin-slice CT to complete CT image registration and applicator reconstruction:

[0060] After preprocessing the CT image of the clinical applicator, the corresponding pipeline data in the applicator template library based on thin-slice CT is matched according to the name of each pipeline of the applicator. Then, a multiple registration method based on improved physical structure-driven optimization is used to complete the three-dimensional spatial registration of the clinical applicator CT image and the corresponding applicator CT image in the applicator template library based on thin-slice CT, and the corresponding relationship between the two applicators in three-dimensional space is established.

[0061] The center point set fitting curve function of the applicator in the thin-layer CT-based applicator template library is called. Based on the above alignment results, the center point set on the clinical applicator CT image is corrected, and the applicator outline is created on the clinical applicator CT image according to the physical size of the applicator to complete the reconstruction of the clinical applicator.

[0062] It should be clarified that the applicator reconstruction of the present invention is a method for extracting and correcting the center point position of the applicator pipe used on the CT image of the metal applicator obtained by the CT image scanning scheme during clinical use.

[0063] The specific implementation scheme of the source reconstruction proposed by the present invention is carried out according to the following steps: Figure 1 The following is a flow chart of the program:

[0064] Establishing a template library of applicators based on thin-slice CT:

[0065] Obtain thin-slice CT data of the applicator in a single canal (including the left oval canal, right oval canal, 15-degree uterine canal, 30-degree uterine canal, 45-degree uterine canal, and 60-degree uterine canal) without metal bolts or other accessories. The CT slice thickness of the present invention is set to 0.6 mm, which can be called thin-slice CT.

[0066] Based on the above thin-slice CT data, the applicator center point set extraction and curve fitting based on hybrid geometric features are respectively used for the applicator. The relevant processes are as follows: Figure 2 As shown. The specific implementation steps are as follows: (1) First, the Gaussian filter noise reduction method is used to preprocess the CT image; (2) Then, the dynamic threshold method is used to extract the three-dimensional center point of the applicator pipe in the thin-layer CT image. Because the applicator pipe has a certain length and the scanning range is generally greater than 5 cm, the number of extracted center points is large. For accurate description, it is called the center point set; (3) Point cloud sorting is carried out along the applicator path, the endpoints of the applicator center point set are identified, and the number of point sets is enhanced at the applicator endpoints; (4) The above center point set is subjected to curve fitting based on the spline function to obtain the fitting curve function; curvature analysis and segmentation are performed based on the above fitting curve function equation; (5) Based on the curve fitting of the geometric morphology of the applicator, the three-segment smooth connection of the curve is enhanced, that is, the geometric morphology characteristics of the straight line segment, the curved segment, and the straight line segment;

[0067] The thin-slice CT images and fitting curve results are saved as XML format files and stored as applicator templates with corresponding names, thereby completing the establishment of an applicator template library based on thin-slice CT.

[0068] Based on the improved sparse cascade network model (Sparse-Cascade Net), rapid identification and accurate classification of clinically used applicators (combinations with assembly accessories such as metal clips), Figure 3 This is a flowchart of the improved sparse cascade network model. The improved sparse cascade network model can quickly (<1s) and accurately (recognition rate>95%) identify the type of applicator combination and uterine canal angle used in clinical practice. The improved sparse cascade network model framework is shown in Figure 4 , the specific implementation process is as follows:

[0069] The data input for the improved sparse cascade network model is CT images of clinical applicators. The dataset used for data modeling and validation in this invention covers the following applicator usage modes: single-tube applicators (15-degree uterine canal, 30-degree uterine canal, 45-degree uterine canal, and 60-degree uterine canal), two-tube applicators (applicators that combine left and right oval canals), and three-tube applicators (applicators that combine left and right oval canals with uterine canals of various angles (15-degree uterine canal, 30-degree uterine canal, 45-degree uterine canal, and 60-degree uterine canal). Therefore, there are a total of nine categories.

[0070] Improved sparse cascade network model Input data first passes through the dynamic sparse activation module, see Figure 5 ,The dynamic sparse activation module generates a weight map by a gated convolution layer (Gated Conv), sets the dynamic threshold to 30%, and outputs the weight map for simple and complex areas respectively;

[0071] For simple regions, a lightweight branch module is used to further extract features, such as Figure 6 The network structure of the lightweight branch module is shown in Figure 2; for complex areas, the deep branch module is used to further extract features, such as Figure 7 The network structure of the deep branch module shown;

[0072] After the features obtained by the lightweight branch module and the features obtained by the deep branch are spliced ​​together, they are passed through a cascade decision maker module. Figure 8 As shown in the figure, the first stage is global average pooling, which is used to obtain confidence, and the second stage is the Transformer encoder block. If the first stage obtains a high confidence level (≥0.85), a fully connected layer is connected to directly give the classification result without going through the second stage. If the confidence level is low (<0.85), the second stage Transformer encoder block is required to complete the final classification and output the classification result.

[0073] Preprocessing of clinical applicator CT images: After the clinical applicator CT images are identified and classified by the above-mentioned trained improved sparse cascade network model, the CT images are further processed using a three-dimensional feature extraction method, including removing noise, foreign matter (image enhancer, contraceptive ring or calcification points, etc.) and splitting the applicator while retaining the physical morphological characteristics of the applicator.

[0074] The three-dimensional feature extraction method is specifically as follows:

[0075] Threshold Adaptation: A dynamic threshold algorithm based on local grayscale statistics, combined with a cascade of morphological closing and opening operations (two iterative closing operations followed by one iterative opening operation), eliminates isolated noise points and other foreign objects caused by metal artifacts while preserving the intact structure of the applicator.

[0076] Acquiring a center point set based on geometric features: This process uses different center point set extraction strategies based on the number of applicator tubes. When the applicator is a single tube, the present invention employs anisotropic morphological filtering: a combination of a 3×3×3 cubic kernel and a 2×2×2 sphere kernel is used, respectively. A curvature-constrained skeleton smoothing technique enforces a curvature change rate threshold (<1.5 cm) to avoid anatomical distortion caused by oversmoothing. When the applicator is multi-tube (two-tube and three-tube applicators), after center point extraction is complete, the multiple tubes are separated to complete the applicator tube segmentation. The left and right oval canals and the uterine canal (three-tube applicators) are identified based on the applicator's physical location. For example, a three-channel applicator with a 45-degree uterine canal and left and right oval canals will be segmented into three sets of CT images. Each set of CT images only retains a single channel of the corresponding applicator. In this example, the three CT images contain only the 45-degree uterine canal, the left oval canal, and the right oval canal. These CT images will be subsequently matched with the applicator data for the corresponding channel in the thin-slice CT-based applicator template library. During the segmentation of a multi-channel applicator, image feature marker parameters tailored to clinical scanning conditions (a 3x3x4, all-ones three-dimensional matrix) are used to extract the maximum connected component. The number of maximum connected components is then limited based on the number of channels to complete the segmentation.

[0077] After the split CT image of the clinical applicator is matched with the corresponding single-channel applicator in the applicator template library based on thin-slice CT, the present invention uses a multiple registration technology based on physical structure driven optimization to complete the spatial position correspondence of the two sets of CT images and finally realize the reconstruction of the clinical applicator. The process can be seen Figure 9 The specific implementation process is as follows:

[0078] Initial image registration: The bone-based registration method is used to complete the approximate registration of the two CT scans;

[0079] The fitting curve obtained from the applicator template library based on thin-slice CT is used as the "shape skeleton" to guide the point set formed by the clinical applicator CT image to fit within the target range;

[0080] The center point set extraction and fitting of the applicator are used clinically to strengthen the three-segment structural characteristics;

[0081] Smoothing based on physical structure and hard constraints on transition points;

[0082] Weighted registration: Strengthen the weight of curved segments and strengthen the registration of curved segments in multi-segment registration, ultimately achieving 3D registration of clinical applicators with corresponding applicators in the thin-slice CT-based applicator template library;

[0083] Based on the constrained fitting of the center point set of the clinical applicator, the contour of the physical dimensions of the pipeline is reconstructed.

[0084] The relevant equations and objective function formulas used in the present invention are:

[0085] (1)

[0086] This formula is a curvature analysis equation used to segment the fitting curve. is the equation of the fitting curve of the three-dimensional center point of the applicator pipeline (in arc length as parameters), and are the first and second derivatives (tangent vector and curvature vector), respectively.

[0087] (2)

[0088] This formula is the objective function of the secondary registration (geometric constraint optimization), which aims to optimize the rigid transformation parameters so that the point set formed by the transformed clinical applicator CT image is consistent with the segmented geometric features of the thin-slice CT image curve ( Line segments and curved segments).

[0089]

[0090] This formula is the energy function of the smoothing spline (used for piecewise constrained fitting), and its purpose is to perform smooth curve fitting under the conditions of known piecewise structure (straight and curved segments) and transition point constraints. It is a fitting curve function in the template library of the applicator based on thin-slice CT. is a parameter. is the weight (more weight at curved segments). is the smoothing coefficient. The second term is the smoothing regularization term (penalizing curvature). The transition points (the start and end points of the curved segments) impose position constraints: .in, and is the position of the thin layer curve at the transition point. The third item is the constraint point limit (strengthening the bending segment split point), is the constraint strength coefficient.

[0091] In summary, the present invention obtains CT data of each pipeline of the applicator through thin-slice CT scanning and fits the curve equation of each pipeline to establish an applicator template library based on thin-slice CT;

[0092] An improved sparse cascade network model for rapid identification and classification was established, fully considering the efficiency and accuracy requirements of clinical work. The improved sparse cascade network model can quickly and accurately classify applicators, providing a basis for subsequent matching of applicator pipelines in the thin-slice CT-based applicator template library.

[0093] Based on thin-slice CT, the CT-CT image registration between the CT data in the applicator template library based on thin-slice CT and the CT data of the clinical applicator is more efficient and faster; using the applicator fitting curve established with thin-slice CT data as the skeleton, segmented fitting, physical constraints and other methods are used to optimize the fitting curve of the central point set extracted from the CT data of the clinically used applicator, thereby completing the precise reconstruction of the applicator.

Claims

1. A method for reconstructing an applicator, characterized in that: The thin-slice CT data of each pipeline of the applicator are acquired through thin-slice CT scanning, the curve equation of each pipeline is fitted, and the template library of the applicator based on thin-slice CT is established; An improved sparse cascade network model was established and trained using CT images of applicators under conventional scanning conditions. This model was used to classify clinical applicators into single or multiple tubes. In the case of multiple tubes, the tubes were split based on the classification results to obtain multiple single tubes. Single tubes were matched to applicators in the applicator template library based on thin-slice CT. The CT images in the thin-slice CT-based applicator template library and the CT images of the clinical applicator are registered using a multiple registration method based on an improved physical structure-driven optimization; using the applicator fitting curve established in the thin-slice CT-based applicator template library as the skeleton, the fitting curve of the central point set extracted from the CT image of the clinical applicator is optimized using segmented fitting and physical constraints to complete the accurate reconstruction of the applicator.

2. The method for reconstructing the applicator according to claim 1, characterized in that: The process of acquiring the template library of the applicator based on thin-slice CT is as follows: obtaining a thin-slice CT image of only the canal portion of the applicator; the canal portion includes the left oval canal, the right oval canal, the 15-degree uterine canal, the 30-degree uterine canal, the 45-degree uterine canal, and the 60-degree uterine canal; Based on the above thin-slice CT images, the center point set extraction and curve fitting of each type of applicator are performed based on hybrid geometric features; the above thin-slice CT images and fitting curve results are stored as an applicator template library based on thin-slice CT.

3. The method for reconstructing the applicator according to claim 2, characterized in that: The specific implementation process of the source center point set extraction and curve fitting based on hybrid geometric features is as follows: The thin-slice CT images were preprocessed using Gaussian filtering denoising method; The dynamic threshold method is used to extract the three-dimensional center point of the applicator pipeline in the thin-slice CT image and obtain the center point set; Sorting the point cloud along the applicator path, identifying the endpoints of the applicator center point set, and enhancing the number of point sets at the applicator endpoints; Performing curve fitting based on a spline function on the above-mentioned central point set to obtain a fitting curve function; Based on the above fitting curve function, curvature analysis and segmentation are performed, and the segmentation equation is as follows; (1) is the partition equation; is the equation of the fitting curve of the three-dimensional center point of the applicator pipeline; is the arc length, and They are The first and second derivatives of ; Based on the curve fitting of the geometric morphology of the applicator, the three-segment smooth connection of the curve is enhanced, and the geometric morphological features of straight line segment, curved segment and straight line segment are obtained in sequence.

4. The method for reconstructing the applicator according to claim 2, wherein: The thin-slice CT image is CT image data with a slice thickness of 0.6 mm.

5. The method for reconstructing the applicator according to claim 1, characterized in that: The improved sparse cascade network model includes a dynamic sparse activation module, a light branch module for processing simple areas, a deep branch module for processing complex areas, and a cascade decision maker module; the data processed by the deep branch module and the light branch module are feature-joined, and the data after feature-joining passes through a cascade decision maker module, and the high-confidence results are directly output, while the low-confidence results are further output after passing through a transformer module; Input the CT image of the clinical applicator. The applicator types include: single-tube applicator: 15-degree uterine canal, 30-degree uterine canal, 45-degree uterine canal, and 60-degree uterine canal; two-tube applicator: an applicator used in combination with the left oval canal and the right oval canal; three-tube applicator: an applicator used in combination with the left oval canal and the right oval canal at various angles of the uterine canal.

6. The method for reconstructing the applicator according to claim 5, characterized in that: The input data of the improved sparse cascade network model first passes through the dynamic sparse activation module; The dynamic sparse activation module generates a mask by a gated convolutional layer with a dynamic threshold set to 30% to identify key areas; In the above process, the mask generated by the gated convolution layer is regarded as a weight map, which is used together with the dynamic threshold to process the input data. After processing the weight map, two data of the same size are output; the weight values ​​are arranged from large to small; the values ​​of the points before 30% of the weight value are retained, and the values ​​of the other points are set to 0, and this area is defined as a complex area; the values ​​of the points after 30% of the weight value are retained, and the values ​​of the other points are set to 0, and this area is defined as a simple area; The simple area output by the dynamic sparse activation module uses a lightweight branch module to extract features; The complex area output by the dynamic sparse activation module is used to extract features using a deep branch module; The features obtained from the lightweight branch module and the features obtained from the deep branch are concatenated and then passed through a cascade decision maker module; the first level of the cascade decision maker module is global average pooling, connected to a fully connected layer, and the second level is a Transformer encoder block; if the confidence level obtained after the first level is ≥0.85, the classification result is given directly without passing through the second level; when the confidence level is <0.85, the final classification is completed through the second-level Transformer encoder block, and the classification result is output.

7. The method for reconstructing the applicator according to claim 1, characterized in that: The registration is specifically as follows: After preprocessing the CT image of the clinical applicator, the corresponding pipeline data in the thin-slice CT-based applicator template library are matched according to the name of each pipeline of the applicator. Then, a multiple registration method based on improved physical structure-driven optimization is used to complete the three-dimensional spatial registration of the CT image of the clinical applicator with the CT image of the corresponding applicator in the thin-slice CT-based applicator template library, and establish the corresponding relationship between the two applicators in three-dimensional space. The reconstruction of the clinical applicator is specifically as follows: calling the center point set fitting curve function of the applicator template library based on thin-layer CT, correcting the center point set on the clinical applicator CT image based on the above-mentioned alignment results, and creating the applicator outline on the clinical applicator CT image according to the physical size of the applicator to complete the reconstruction of the clinical applicator.

8. The method for reconstructing the applicator according to claim 7, characterized in that: The specific implementation process of the multiple registration method based on improved physical structure driven optimization is as follows: Initial image registration: The bone-based registration method is used to complete the approximate registration of the CT image of the corresponding applicator in the thin-slice CT-based applicator template library with the clinical applicator CT image; The fitting curve obtained from the applicator template library based on thin-slice CT is used as the "shape skeleton" to guide the point set formed by the clinical applicator CT image to fit within the target range; Extraction and fitting of the center point set of the clinical applicator to strengthen the three-segment structural features; the three-segment structural features are straight segment, curved segment, and straight segment in sequence; Weighted registration: The weight of the curved segment is enhanced, and the registration of the curved segment is strengthened in the multi-segment registration, and finally the three-dimensional spatial registration of the clinical applicator and the corresponding applicator in the applicator template library based on thin-slice CT is achieved. The objective function of the relevant registration is given by formula (2); The physical dimensions of the pipeline are reconstructed through constrained fitting of the center point set of the clinical applicator. （2） in, is the constraint weight of the line segment, It represents the distance between the point corresponding to the straight line segment extracted from the clinical applicator CT image after displacement and the straight line segment in the CT image of the applicator template library based on thin-slice CT. is the corresponding point in the straight line segment after displacement, is a straight line segment in thin-slice CT, is the constraint weight of the curved segment, is the corresponding point in the curved line segment after displacement, It is the corresponding point set in the curved segment of the CT image of the applicator template library based on thin-slice CT; It represents the sum of the distances from the point corresponding to the straight line segment extracted from the clinical applicator CT image after displacement to the straight line segment in the CT image of the applicator template library based on thin-slice CT. represents the sum of the minimum values ​​of the distances between the points corresponding to the curved segments extracted from the clinical applicator CT image after displacement and the curved segments in the CT images of the applicator template library based on thin-slice CT; Formula (2) is the objective function of weighted registration, which aims to optimize the rigid transformation parameters so that the center point set extracted from the clinical applicator CT image is aligned with the segmented geometric features of the fitting curve corresponding to the applicator template library based on thin-slice CT; （3） Formula (3) is the energy function of the smoothing spline, which is used for piecewise constrained fitting. Under the condition of known piecewise structure and transition point constraints, smooth curve fitting is performed; in: It is a fitting curve function in the template library of the applicator based on thin-slice CT. The fitting function of the center point set of the clinical applicator CT image, is a parameter; is the weight, which is greater at the bend; is the smoothing coefficient; The center point of the clinical applicator CT image is located in the straight line segment. Indicates that the center points in the applicator template library based on thin-slice CT are concentrated at the intersection of the straight line segment and the curve segment, The function representing the center point set of the clinical applicator CT image at the intersection of the straight line-curve and the curve-straight line, is the constraint strength coefficient; Apply position constraints at the transition points: in, and It is the position of the transition point of the fitting curve of the applicator template library based on thin-slice CT.

Images