An automatic positioning method for a pelvic implant channel

Abstract

The application discloses a kind of automatic positioning methods of pelvic implant passage, comprising the following steps: step 1, establish pelvic general anatomical three-dimensional template;Step 2, define implant passage template based on the three-dimensional template of step 1;Step 3, registration is implemented to non-homologous pelvic CT image based on internal information;Step 4, obtain the coordinate conversion between CT coordinate system and anatomical coordinate system;Step 5, the alignment between target pelvic in non-homologous pelvic and pelvic general anatomical three-dimensional model is implemented;Step 6, based on step 2 and step 5, establish passage automatic search.The application adopts the method of three-dimensional model statistical analysis to obtain accurate pelvic general anatomical three-dimensional template, realizes the alignment between different patient models using image registration, based on the space transformation between image coordinate system and anatomical coordinate system, the pelvic implant passage of each patient can be obtained by the method of three-dimensional model registration, so as to shorten the operation time, guide the operation of doctor, and reduce the purpose of doctor's manual operation.

Description

Technical Field

[0001] This invention relates to the field of orthopedic surgical planning technology, and in particular to an automatic positioning method for pelvic implant channels. Background Technology

[0002] Surgical planning is the process of obtaining medical images (such as CT / MRI) of the patient's lesion before surgery, and combining these with the doctor's knowledge of anatomy and pathology to determine, for example, the surgical method, surgical procedure, surgical incision and path, and to obtain a surgical plan.

[0003] In pelvic implant surgery, surgical planning requires determining the implant channel. During surgery, medical instruments such as steel nails and Kirschner wire screws are inserted into the determined implant channel to fix the fracture site. The implantation effect not only affects the surgical outcome but also plays a crucial role in postoperative recovery. Current methods for determining the implant first require the doctor to make a subjective judgment based on medical images, then conduct manual experiments using cadaver bone or 3D printing to determine the approximate location of the implant channel, and finally determine the surgical plan.

[0004] Existing methods suffer from a high degree of subjectivity, relying too heavily on preoperative image interpretation and surgeon experience. This leads to a series of problems, including excessive subjectivity and difficulty in manual operation. Compared with traditional preoperative planning for pelvic surgery, the method of selecting a three-dimensional model for statistical analysis can obtain a precise three-dimensional template of universal pelvic anatomy. Image registration is used to align different patient models. Based on the spatial transformation between the image coordinate system and the anatomical coordinate system, the three-dimensional model registration method can automatically obtain the pelvic implant channel for each patient, saving doctors' time, simplifying their operation, and achieving a more objective and accurate method for establishing implant channels. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide an automatic positioning method for pelvic implant channels. By utilizing internal bone information and image registration, the channel of the patient during surgery is automatically determined based on the patient's medical image data, thereby shortening the operation time, guiding the doctor's operation, and reducing the doctor's manual operation.

[0006] To address the aforementioned technical problems, this invention provides an automatic positioning method for pelvic implant channels, comprising the following steps:

[0007] Step 1: Create a three-dimensional template for general pelvic anatomy;

[0008] Step 2: Define the implant channel template based on the 3D template from Step 1;

[0009] Step 3: Register non-homogeneous pelvic CT images based on internal information;

[0010] Step 4: Obtain the coordinate transformation between the CT coordinate system and the anatomical coordinate system;

[0011] Step 5: Align the target pelvis with the general anatomical 3D model of the pelvis in non-homogeneous pelvises;

[0012] Step 6: Establish automatic channel search based on Step 2 and Step 5.

[0013] Preferably, step 1, establishing a general anatomical three-dimensional template for the pelvis, specifically includes the following steps:

[0014] Step 11: Construct a pelvic model library; collect pelvic models of people of different ages and genders, and name the collection of these models the pelvic model library;

[0015] Step 12: Generate the average 3D model; using models from the pelvic model library as input, read in the point and surface information of all models, and apply a statistical morphological model filter to perform statistical morphological analysis on the point and surface information of all models to obtain the morphological error E. all =e r {r=1,2,..n}, after analysis, all models are used to fit the predicted statistical model. Based on the point and surface data and error of the fitted statistical morphological model output by the statistical morphological model filter, the average statistical morphological model M is constructed using the statistical morphological model builder. SSM .

[0016] Preferably, in step 2, the implant channel template is defined based on the three-dimensional template in step 1, and based on clinical experience, the average statistical morphological model M of the pelvis is used. SSM Define pelvic implant channel C SSM This serves as a template for automatically exporting pelvic implant channels for other patients.

[0017] Preferably, step 3, which involves registering non-homogeneous pelvic CT images based on internal information, specifically includes the following steps:

[0018] Step 31: Perform preprocessing of the CT data corresponding to the model library;

[0019] Step 32: The target image for registration is a CT image of any randomly selected patient, and the image to be registered is a CT image of the remaining patients. The registration network includes some encoder-decoder combinations containing skip connections and several consecutive layer combinations.

[0020] Preferably, in step 31, the preprocessing of the CT data corresponding to the model library is specifically as follows: first, the CT body data containing the entire lower body is used as input, then the cross section of the region of interest, namely the pelvic region, is automatically extracted, and finally the extracted cross section is used to compress the CT body data to 440*256*256 for subsequent processing.

[0021] Preferably, in step 32, the encoding operation adopted in the encoder is an image pyramid structure, using a three-dimensional convolution with a kernel size of 3x3x3 and a stride of 2. In terms of representation dimension, the spatial dimension is reduced to 1 / 2 of the original dimension each time under the action of the convolution stride, until it is reduced to 1 / 16 of the original dimension.

[0022] Preferably, in step 32, upsampling, convolution, and connection skipping are used alternately in the decoder; in subsequent successive layers, the receptive field of the image is refined, thereby achieving more accurate alignment of anatomical structures, and the spatial registration deformation field from the image to be registered to the target image is output after network registration.

[0023] Preferably, in step 4, obtaining the coordinate transformation between the CT coordinate system and the anatomical coordinate system specifically involves: outputting... This refers to the variation of each voxel in the CT volume data in three orientations within the image coordinate system, while the automatically established channel is located inside the 3D model in the anatomical coordinate system. First, the affine matrix T is extracted from the CT volume data, and then the deformation field of S2 is output. As input, the two are multiplied together to obtain a spatial transformation in anatomical coordinates.

[0024] Preferably, in step 5, aligning the target pelvis with the universal anatomical 3D model of the pelvis in non-homogeneous pelvis specifically involves: aligning the image to be registered with the target image V... i and V fixed Inputting this module converts the data into a corresponding 3D model S based on its voxel information. i and S fixed .

[0025] Preferably, in step 6, the automatic channel search based on steps 2 and 5 specifically involves: setting M... SSM and its corresponding channel C SSM As input, M SSM and target model S fixed Registration yields the transformation matrix T. trans According to channel C SSM In M SSM The positional relationship in the model leads to the target model S. fixed Channel C fixed Then the 3D model S to be registered i and target model S fixed V i To V fixed Spatial transformation Input, target 3D model S fixed The passage above All input modules, then spatial transformation Acting on S i Obtain the transformed 3D model S i Registration to S fixed Finally, according to and S fixed The correspondence, using channels Solve for the channels of the registered model For S i and Transformation relationship between Find the inverse, and then apply the result to... This will output the corresponding channels from the original model. This completes the automatic search and establishment of the channel.

[0026] The beneficial effects of this invention are as follows: This invention can obtain a precise universal pelvic anatomical 3D template by adopting a 3D model statistical analysis method, and achieve alignment between different patient models by using image registration. Based on the spatial transformation between the image coordinate system and the anatomical coordinate system, the pelvic implant channel of each patient can be automatically obtained through the 3D model registration method, thereby shortening the operation time, guiding the doctor's operation, and reducing the doctor's manual operation. Attached Figure Description

[0027] Figure 1 This is a schematic diagram of the method flow of the present invention.

[0028] Figure 2 This is a schematic diagram of the network structure of the image registration module of the present invention. Detailed Implementation

[0029] like Figure 1 As shown, an automatic positioning method for pelvic implant channels includes the following steps:

[0030] Step 1: Create a three-dimensional template for general pelvic anatomy; specifically including the following steps:

[0031] Step 11: Construct a pelvic model library; collect pelvic models of people of different ages and genders, and name the collection of these models the pelvic model library;

[0032] Step 12: Generate the average 3D model; using models from the pelvic model library as input, read in the point and surface information of all models, and apply a statistical morphological model filter to perform statistical morphological analysis on the point and surface information of all models to obtain the morphological error E. all =e r{r=1,2,..n}, after analysis, all models are used to fit the predicted statistical model. Based on the point and surface data and error of the fitted statistical morphological model output by the statistical morphological model filter, the average statistical morphological model M is constructed using the statistical morphological model builder. SSM .

[0033] Step 2: Define the implant channel template based on the 3D template from Step 1; based on clinical experience, in the pelvic average statistical morphological model M... SSM Define pelvic implant channel C SSM This serves as a template for automatically exporting pelvic implant channels for other patients.

[0034] Step 3: Register non-homogeneous pelvic CT images based on internal information; such as... Figure 2 As shown, the specific steps include the following:

[0035] Step 31: Perform preprocessing of the CT data corresponding to the model library. First, take the CT body data containing the entire lower body as input, then automatically extract the section of the region of interest, namely the pelvic region, and finally use the extracted section to compress the CT body data to 440*256*256 for subsequent processing.

[0036] Step 32: The target image for registration is a randomly selected CT image of any patient, while the images to be registered are CT images of other patients. The registration network includes encoder-decoder combinations with skip connections and several consecutive layers. In the encoder, the encoding operation adopts an image pyramid structure, using a 3x3x3 three-dimensional convolution with a stride of 2. In terms of representation dimension, the spatial dimension is reduced to 1 / 2 of its original size each time under the effect of the convolution stride, until it is reduced to 1 / 16 of its original dimension. In the decoder, upsampling, convolution, and connection skips are used alternately. In subsequent consecutive layers, the receptive field of the image is refined, thereby achieving more accurate alignment of anatomical structures. After network registration, the output is the spatial registration deformation field from the image to the target image.

[0037] Step 4: Obtain the coordinate transformation between the CT coordinate system and the anatomical coordinate system; output This refers to the variation of each voxel in the CT volume data in three orientations under the image coordinate system, while the automatically established channel is located inside the 3D model in the anatomical coordinate system. Therefore, we first extract the affine matrix T from the CT volume data, and then output the deformation field of S2. As input, the two are multiplied together to obtain a spatial transformation in anatomical coordinates.

[0038] Step 5: Align the target pelvis with the universal pelvic anatomical 3D model in the non-homogeneous pelvis; connect the image to be registered and the target image V. i and V fixed Inputting this module converts the data into a corresponding 3D model S based on its voxel information. i and S fixed .

[0039] Step 6: Establish automatic channel search based on Steps 2 and 5; set M SSM and its corresponding channel C SSM As input, M SSM and target model S fixed Registration yields the transformation matrix R. trans According to channel C SSM In M SSM The positional relationship in the model leads to the target model S. fixed Channel C fixed Then the 3D model S to be registered i and target model S fixed V i To V fixed Spatial transformation Input, target 3D model S fixed The passage above All input modules, then spatial transformation Acting on S i Obtain the transformed 3D model S i Registration to S fixed Finally, according to and S fixed The correspondence, using channels Solve for the channels of the registered model For S i and Transformation relationship between Find the inverse, and then apply the result to... This will output the corresponding channels from the original model. This completes the automatic search and establishment of the channel.

Claims

1. An automatic positioning method for pelvic implant channels, characterized in that, Includes the following steps: Step 1: Create a three-dimensional template for general pelvic anatomy; Step 2: Define the implant channel template based on the 3D template from Step 1; Step 3: Register non-homogeneous pelvic CT images based on internal information; specifically including the following steps: Step 31: Perform preprocessing of the CT data corresponding to the model library. First, take the CT body data containing the entire lower body as input, then automatically extract the section of the region of interest, namely the pelvic region, and finally use the extracted section to compress the CT body data to 440*256*256 for subsequent processing. Step 32: The target image for registration is a randomly selected CT image of any patient, while the images to be registered are CT images of other patients. The registration network includes encoder-decoder combinations with skip connections and several consecutive layers. In the encoder, the encoding operation adopts an image pyramid structure, using a 3x3x3 three-dimensional convolution with a stride of 2. In terms of representation dimension, the spatial dimension is reduced to 1 / 2 of its original size each time under the effect of the convolution stride, until it is reduced to 1 / 16 of its original dimension. In the decoder, upsampling, convolution, and connection skips are used alternately. In subsequent consecutive layers, the receptive field of the image is refined, thereby achieving more accurate alignment of anatomical structures. After network registration, the output is the spatial registration deformation field from the image to the target image. Step 4: Obtain the coordinate transformation between the CT coordinate system and the anatomical coordinate system; Step 5: Align the target pelvis with the general anatomical 3D model of the pelvis in non-homogeneous pelvises; Step 6: Establish automatic channel search based on Step 2 and Step 5.

2. The automatic positioning method for pelvic implant channels as described in claim 1, characterized in that, Step 1, establishing a general 3D pelvic anatomy template, specifically includes the following steps: Step 11: Construct a pelvic model library; collect pelvic models of people of different ages and genders, and name the collection of these models the pelvic model library; Step 12: Generate the average 3D model; using models from the pelvic model library as input, read in the point and surface information of all models, and apply a statistical morphological model filter to perform statistical morphological analysis on the point and surface information of all models to obtain the morphological error E. all =e r {r=1,2,..n}, after analysis, all models are used to fit the predicted statistical model. Based on the point and surface data and error of the fitted statistical morphological model output by the statistical morphological model filter, the average statistical morphological model M is constructed using the statistical morphological model builder. SSM .

3. The automatic positioning method for pelvic implant channels as described in claim 1, characterized in that, In step 2, an implant channel template is defined based on the three-dimensional template from step 1. Based on clinical experience, a statistical morphological model M is used. SSM Define pelvic implant channel C SSM This serves as a template for automatically exporting pelvic implant channels for other patients.

4. The automatic positioning method for pelvic implant channels as described in claim 1, characterized in that, In step 4, obtaining the coordinate transformation between the CT coordinate system and the anatomical coordinate system specifically involves: Output This refers to the variation of each voxel in the CT volume data in three orientations within the image coordinate system, while the automatically established channel is located inside the 3D model in the anatomical coordinate system. First, the affine matrix T is extracted from the CT volume data, and then the deformation field of S32 is output. As input, the two are multiplied together to obtain a spatial transformation in anatomical coordinates.

5. The automatic positioning method for pelvic implant channels as described in claim 1, characterized in that, In step 5, the alignment between the target pelvis and the universal pelvic anatomical 3D model in non-homogeneous pelvises is specifically performed by: aligning the image V to be registered... i and target image V fixed The input is converted into a corresponding 3D pelvic model S to be registered based on its voxel information. i and the target pelvic 3D model S fixed .

6. The automatic positioning method for pelvic implant channels as described in claim 1, characterized in that, In step 6, the automatic channel search based on steps 2 and 5 specifically involves: establishing the statistical morphological model M... SSM and its corresponding pelvic implant channel C SSM As input, the statistical morphological model M SSM and the target pelvic 3D model S fixed Registration yields the transformation matrix T. trans According to the pelvic implant channel C SSM In the statistical morphological model M SSM The positional relationship in the model leads to the 3D model S of the target pelvis. fixed Channel C fixed Then, the 3D model of the pelvis to be registered, S i and the target pelvic 3D model S fixed The image to be registered, V i To target image V fixed Spatial transformation Input: Target pelvic 3D model S fixed The passage above All inputs are then processed, followed by spatial transformation. The three-dimensional pelvic model S to be registered i Obtain the transformed 3D model The 3D model of the pelvis to be registered S i Registered to the target pelvic 3D model S fixed Finally, based on the transformed 3D model and the target pelvic 3D model S fixed The correspondence, using channels Solve for the channels of the registered model The 3D model of the pelvis to be registered i With the transformed 3D model Transformation relationship between Inverse the algorithm and then apply the result to the channels of the registered model. Output the corresponding channels of the original model This completes the automatic search and establishment of the channel.

Images