Preoperative planning method and system based on three-dimensional ankle joint replacement CT image

Abstract

The invention discloses a preoperative planning method and system based on a three-dimensional ankle joint replacement CT image, and the method comprises the steps: carrying out the data processing of to-be-processed ankle joint image data through a statistical model, and constructing a three-dimensional average ankle joint point cloud model; registering the three-dimensional average ankle joint point cloud model with ankle joint image data to be processed through a CPD non-rigid point cloud registration algorithm to obtain an ankle joint prosthesis placement position; performing data prediction on the to-be-processed ankle joint image data through a random forest algorithm to obtain ankle joint prosthesis type data; and three-dimensional ankle joint replacement preoperative planning is realized by combining the ankle joint prosthesis placement position and the ankle joint prosthesis body size data. According to the method, the accurate ankle joint registration result can be achieved in the three-dimensional space by reducing the difference between individuals. The preoperative planning method and system based on the three-dimensional ankle joint replacement CT image can be widely applied to the technical field of ankle joint surgical planning.

Description

Technical Field

[0001] The present invention relates to the technical field of ankle joint surgical planning, and particularly to a preoperative planning method and system based on three-dimensional ankle joint replacement CT images. Background Art

[0002] Ankle joint surgical planning is for the preoperative planning of ankle joint replacement surgery. Before excising the damaged part of the ankle joint, preoperative surgical planning should be carried out on the patient's ankle joint. The purpose of this surgical planning is to quickly find the corresponding prosthesis placement position and the model of the prosthesis (the part that needs to be cut) during the operation. Ankle joint surgical planning can help doctors quickly find the incision position and the size of the bone to be cut during the entire process of ankle joint replacement surgery. The preoperative planning task of the ankle joint in the related art is usually carried out under two-dimensional images, and protective measures for blood vessels and nerves during the operation are considered. Among them, in the traditional method of the prosthesis implantation plan, two-dimensional X-ray films are marked. The main purpose of the marking is to determine the mechanical axis of the tibia. The mechanical axis of the tibia is the line connecting the midpoint of the intercondylar ridge of the tibia and the center of the ankle joint. The center of the ankle joint is taken as the midpoint of the medial and lateral malleoli. The prosthesis position of the ankle joint can be determined according to the mechanical axis. The upper two vertices of the talus also need to be marked to facilitate the determination of the prosthesis model. In this way, the traditional surgical planning will provide a relatively clear surgical idea and guidance for doctors during the operation. However, surgical planning on two-dimensional X-rays often cannot show the three-dimensional characteristics of the tibia, and some information will be lost during surgical planning. And the results obtained through two-dimensional preoperative planning are not very clear corresponding to the real human tibia during the operation. Currently, medical images are used for processing to obtain a three-dimensional bone model, and simulation installation experiments are carried out in three-dimensional space. However, it only performs simulated surgery in three-dimensional space, does not have intelligence, or does not have a specific surgical planning process, but is a method of three-dimensional reconstruction and effect enhancement. Summary of the Invention

[0003] In order to solve the above technical problems, the purpose of the present invention is to provide a preoperative planning method and system based on three-dimensional ankle joint replacement CT images, which can achieve accurate ankle joint registration results in three-dimensional space by reducing the differences between individuals and capturing the non-linear deformation of complex model data.

[0004] The first technical solution adopted by the present invention is: A preoperative planning method based on three-dimensional ankle joint replacement CT images, comprising the following steps:

[0005] Data process the ankle joint image data to be processed through a statistical model, and construct a three-dimensional average ankle joint point cloud model;

[0006] Register the three-dimensional average ankle joint point cloud model with the ankle joint image data to be processed through the CPD non-rigid point cloud registration algorithm to obtain the placement position of the ankle joint prosthesis;

[0007] Perform data prediction on the ankle joint image data to be processed through the random forest algorithm to obtain the ankle joint prosthesis model data;

[0008] Combine the placement position of the ankle joint prosthesis with the ankle joint prosthesis model data to achieve preoperative planning for three-dimensional ankle joint replacement.

[0009] Further, the step of processing the ankle joint image data to be processed through a statistical model to construct a three-dimensional average ankle joint point cloud model specifically includes:

[0010] Obtain the ankle joint image data to be processed;

[0011] Perform modeling processing on the ankle joint image data to be processed through a statistical model to obtain a three-dimensional average ankle joint model;

[0012] Perform data annotation processing on the three-dimensional average ankle joint model to obtain a three-dimensional average ankle joint model with annotations;

[0013] Perform conversion processing on the three-dimensional average ankle joint model with annotations through the average grid sampling algorithm to obtain a three-dimensional average ankle joint point cloud model.

[0014] Further, the step of performing modeling processing on the ankle joint image data to be processed through a statistical model to obtain a three-dimensional average ankle joint model specifically includes:

[0015] Extract data from the ankle joint image data to be processed to obtain a number of distal tibia sample data;

[0016] Perform rigid alignment on a number of distal tibia sample data to obtain aligned distal tibia sample data;

[0017] Perform data normalization processing on the aligned distal tibia sample data to obtain normalized distal tibia sample data;

[0018] Extract feature points from the normalized distal tibia sample data to obtain distal tibia sample feature data;

[0019] Perform principal component analysis on the distal tibia sample feature data to obtain the principal component feature vector;

[0020] Represent the shape of the tibia as the mean shape and combine it with the principal component feature vector to construct a three-dimensional average ankle joint model.

[0021] Further, the step of registering the three-dimensional average ankle joint point cloud model with the ankle joint image data to be processed by the CPD non-rigid point cloud registration algorithm to obtain the placement position of the ankle joint prosthesis specifically includes:

[0022] The ankle joint image data to be processed is transformed by the average grid sampling algorithm to obtain the ankle joint three-dimensional point cloud image data;

[0023] The ankle joint three-dimensional point cloud image data and the three-dimensional average ankle joint point cloud model are loaded into the same space and radius filtering processing is performed to obtain the target point cloud data and the template point cloud model. Among them, the template point cloud model is derived from the three-dimensional average ankle joint point cloud model, and the target point cloud data is derived from the ankle joint three-dimensional point cloud image data;

[0024] The target point cloud data and the template point cloud model are registered by the CPD non-rigid point cloud registration algorithm to obtain the placement position of the ankle joint prosthesis.

[0025] Further, the step of registering the target point cloud data and the template point cloud model by the CPD non-rigid point cloud registration algorithm to obtain the placement position of the ankle joint prosthesis specifically includes:

[0026] Based on the target point cloud data, a Gaussian mixture model sample is constructed;

[0027] The parameters of the Gaussian mixture model sample and the transformation matrix of the template point cloud model are optimized by the EM algorithm, and the optimized transformation matrix is output;

[0028] The optimized transformation matrix is applied to the template point cloud model for registration to obtain the placement position of the ankle joint prosthesis.

[0029] Further, the step of optimizing the parameters of the Gaussian mixture model sample and the transformation matrix of the template point cloud model by the EM algorithm to output the optimized transformation matrix specifically includes:

[0030] The expectation calculation is performed on the Gaussian mixture model sample to obtain the probability distribution between the target point cloud data and the template point cloud model;

[0031] The maximization calculation is performed on the Gaussian mixture model sample to optimize the transformation matrix of the template point cloud model so that the probability distribution of the target point cloud data is maximized on the template point cloud model;

[0032] According to the non-rigid registration mode, the transformation matrix is updated;

[0033] Repeat the step of optimizing the parameters of the Gaussian mixture model sample and the transformation matrix of the template point cloud model by the EM algorithm until the preset number of iterations is satisfied, and the optimized transformation matrix is output.

[0034] Further, the step of performing data prediction on the ankle joint image data to be processed through the random forest algorithm to obtain the ankle joint prosthesis model data specifically includes:

[0035] Annotate and measure the ankle joint image data to be processed to obtain the key feature parameters of the ankle joint image. The key feature parameters of the ankle joint image include the longest distance of the distal tibia, the thickness of the medial malleolus of the distal tibia, the surface area of the distal tibia, and the volume of the distal tibia;

[0036] Perform data cleaning on the key feature parameters of the ankle joint image to obtain the cleaned key feature parameters of the ankle joint image;

[0037] Perform sampling training with replacement on the cleaned key feature parameters of the ankle joint image through the random forest algorithm, and output the ankle joint prosthesis model data.

[0038] The second technical solution adopted by the present invention is: A preoperative planning system based on three-dimensional ankle replacement CT images, including:

[0039] The first module is used to perform data processing on the ankle joint image data to be processed through a statistical model to construct a three-dimensional average ankle joint point cloud model;

[0040] The second module is used to register the three-dimensional average ankle joint point cloud model with the ankle joint image data to be processed through the CPD non-rigid point cloud registration algorithm to obtain the placement position of the ankle joint prosthesis;

[0041] The third module is used to perform data prediction on the ankle joint image data to be processed through the random forest algorithm to obtain the ankle joint prosthesis model data;

[0042] The fourth module is used to combine the placement position of the ankle joint prosthesis with the ankle joint prosthesis model data to achieve preoperative planning for three-dimensional ankle replacement.

[0043] The beneficial effects of the method and system of the present invention are: The present invention performs data processing on the ankle joint image data to be processed through a statistical model, can reduce the differences between individuals through this model to obtain the final average model, and further registers the three-dimensional average ankle joint point cloud model with the ankle joint image data to be processed through the CPD non-rigid point cloud registration algorithm, which can be applied to rigid and non-rigid point cloud registration tasks, and can capture complex non-linear deformations during application. And during registration, no additional manual marking is required, reducing human interference and ensuring the authenticity of the results. Subsequently, data prediction is performed on the ankle joint image data to be processed through the random forest algorithm, and finally, by combining the placement position of the ankle joint prosthesis with the ankle joint prosthesis model data, preoperative planning for three-dimensional ankle replacement is achieved, and accurate ankle joint registration results can be realized in three-dimensional space. Brief Description of the Drawings

[0044] Figure 1 is a flowchart of the steps of a preoperative planning method for a three-dimensional ankle joint replacement CT image according to the present invention;

[0045] Figure 2 is a structural block diagram of a preoperative planning system for a three-dimensional ankle joint replacement CT image according to the present invention;

[0046] Figure 3 is a schematic flow diagram of a three-dimensional ankle joint replacement provided by a specific embodiment of the present invention;

[0047] Figure 4 is a schematic diagram of ankle joint image data to be processed provided by a specific embodiment of the present invention;

[0048] Figure 5 is a schematic flow diagram of data processing of ankle joint image data to be processed through a statistical model provided by a specific embodiment of the present invention;

[0049] Figure 6 is a schematic diagram of a three-dimensional average ankle joint model provided by a specific embodiment of the present invention;

[0050] Figure 7 is a schematic diagram of a three-dimensional average ankle joint model with annotations provided by a specific embodiment of the present invention;

[0051] Figure 8 is a schematic diagram of a three-dimensional average ankle joint point cloud model provided by a specific embodiment of the present invention;

[0052] Figure 9 is a schematic diagram of ankle joint three-dimensional point cloud image data of tibial distal sample data provided by a specific embodiment of the present invention;

[0053] Figure 10 is a schematic flow diagram of the CPD non-rigid point cloud registration algorithm provided by a specific embodiment of the present invention;

[0054] Figure 11 is a schematic diagram of registration result data provided by a specific embodiment of the present invention;

[0055] Figure 12 is a simulation schematic diagram of the registration result provided by a specific embodiment of the present invention;

[0056] Figure 13 is a schematic diagram of key feature parameters of an ankle joint image provided by a specific embodiment of the present invention. Detailed Embodiments

[0057] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. For the step numbers in the following embodiments, they are only set for the convenience of elaboration and explanation, and no limitation is imposed on the order between the steps. The execution order of each step in the embodiments can be adaptively adjusted according to the understanding of those skilled in the art.

[0058] For example, in the three-dimensional reconstruction method and system of the CT image of the bone joint replacement surgery robot proposed in Chinese Patent CN114041878A, first, the bone segmentation data of the CT image is obtained; then, a bone three-dimensional model is generated based on the obtained bone segmentation data of the CT image; an improved SURF feature point video image matching algorithm is used to match and track the target of the video image, then the spatial pose of the target is solved, and then the corresponding spatial transformation is performed on the virtual three-dimensional object to achieve the effect of augmented reality. However, this method does not have a specific surgical planning process, but is a method of three-dimensional reconstruction and effect enhancement.

[0059] Based on this, it should be noted that as Figure 3 shown, the embodiments of the present invention are based on technologies such as three-dimensional reconstruction, statistical shape models, and point cloud registration, select the prosthesis position for the ankle joint CT image, and use technologies such as machine learning and random forest to obtain rich data by measuring the sizes of various types of ankle joints, and perform regression prediction on the newly obtained ankle joint through the data to obtain the corresponding model of the prosthesis. Through the above two processes, the result of the surgical planning can be obtained, and the position where the prosthesis is placed and the corresponding prosthesis model can be determined.

[0060] Referring to Figure 1 , the present invention provides a preoperative planning method based on three-dimensional ankle joint replacement CT images, and the method includes the following steps:

[0061] S100. Process the ankle joint image data to be processed through a statistical model, and construct a three-dimensional average ankle joint point cloud model;

[0062] Specifically, obtain the ankle joint image data to be processed; perform modeling processing on the ankle joint image data to be processed through a statistical model to obtain a three-dimensional average ankle joint model; perform data annotation processing on the three-dimensional average ankle joint model to obtain a three-dimensional average ankle joint model with annotations, as Figure 7 shown; perform conversion processing on the three-dimensional average ankle joint model with annotations through an average grid sampling algorithm to obtain a three-dimensional average ankle joint point cloud model, as Figure 8 shown.

[0063] Among them, the data flow of the statistical model is to extract data from the ankle joint image data to be processed to obtain a number of distal tibia sample data; perform rigid alignment on the number of distal tibia sample data to obtain the aligned distal tibia sample data; perform data normalization processing on the aligned distal tibia sample data to obtain the normalized distal tibia sample data; extract feature points from the normalized distal tibia sample data to obtain distal tibia sample feature data; perform principal component analysis on the distal tibia sample feature data to obtain the principal component feature vector; represent the shape of the tibia as the mean shape and combine it with the principal component feature vector to construct a three-dimensional average ankle joint model.

[0064] In this embodiment, first, read the ankle joint image to be operated on as the input image, as Figure 4 shown; further use the statistical shape model to model the input three-dimensional ankle joint model, and obtain the average model of the input model according to the results of the modeling. This average model has the common features of the input model, and the obtained shape is also similar to the input model; then let an experienced doctor annotate the obtained three-dimensional model to obtain the position where the prosthesis should be placed; finally, convert the annotated average ankle joint model from a three-dimensional surface rendering file to a three-dimensional point cloud file using the average mesh sampling algorithm.

[0065] In some specific embodiments, such as Figure 5As shown, first, 13 distal tibia data are obtained based on the ankle joint image data to be processed. These distal tibia data should be representative and cover most of the possible variations of the distal tibia. Then these tibias are imported into the software. Next, alignment operations are performed. Rigid alignment is carried out on all samples, which includes translation, rotation, and scaling. The center points of the outer bounding boxes of each file are used as the coordinate origins to modify the above three attributes, eliminating the spatial differences between the shapes of these files. The ICP point cloud registration method is used during this process to align these files. Next, these samples are normalized to be within the same scale range to avoid the influence of scale differences. Next, shape representation and feature extraction are carried out. Corresponding feature points (key points) are manually or automatically marked on each sample shape to ensure that the order and quantity of the points are consistent. The selection of feature points should be able to accurately describe the geometry of the shape. Furthermore, the principal component analysis method is used to select the main components that can express these structural and geometric features. First, find all the feature points in all samples, calculate the mean shape of the sample shapes based on these feature points, construct the covariance matrix, centralize the shape data (subtract the mean shape), calculate the covariance matrix, perform eigenvalue decomposition on the covariance matrix to obtain eigenvalues a and eigenvectors b, and finally sort according to the size of the eigenvalues and the eigenvectors, and select the first 32 principal components with a cumulative contribution rate reaching 95%. Finally, model construction is carried out. The shape is represented as a linear combination of the mean shape and the principal components, thus obtaining the average model, as Figure 6 shown.

[0066] Furthermore, after obtaining the average model, first read the triangular mesh model in open3D, use the sample_points_uniformly algorithm for average point sampling and set the downsampling parameter to 1000. Finally, output this file and save it in the ply format.

[0067] S200. Register the three-dimensional average ankle joint point cloud model with the ankle joint image data to be processed through the CPD non-rigid point cloud registration algorithm to obtain the placement position of the ankle joint prosthesis;

[0068] Specifically, the ankle joint image data to be processed is converted through the average grid sampling algorithm to obtain the ankle joint three-dimensional point cloud image data; the ankle joint three-dimensional point cloud image data and the three-dimensional average ankle joint point cloud model are loaded into the same space and radius filtering processing is carried out to obtain the target point cloud data and the template point cloud model. Among them, the template point cloud model is derived from the three-dimensional average ankle joint point cloud model, and the target point cloud data is derived from the ankle joint three-dimensional point cloud image data; the target point cloud data and the template point cloud model are registered through the CPD non-rigid point cloud registration algorithm to obtain the placement position of the ankle joint prosthesis.

[0069] In this embodiment, first, the three-dimensional ankle joint model of the patient is converted from a three-dimensional surface rendering file to a three-dimensional point cloud file using the average mesh sampling algorithm, as Figure 9 shown. Then, the labeled average template is registered with the patient's ankle joint using the CPD non-rigid point cloud registration algorithm to obtain the registered average tibia template. The annotations on the deformed template indicate the correct prosthesis placement positions, as Figure 11 well as Figure 12 shown.

[0070] Among them, as Figure 10 shown, the data processing flow of the CPD non-rigid point cloud registration algorithm is to construct a Gaussian mixture model sample based on the target point cloud data; optimize the parameters of the Gaussian mixture model sample and the transformation matrix of the template point cloud model through the EM algorithm, and output the optimized transformation matrix; apply the optimized transformation matrix to the template point cloud model for registration to obtain the ankle joint prosthesis placement position.

[0071] Furthermore, it should be noted that the data processing flow of the EM algorithm is to perform an expectation calculation on the Gaussian mixture model sample to obtain the probability distribution between the target point cloud data and the template point cloud model; perform a maximization calculation on the Gaussian mixture model sample to optimize the transformation matrix of the template point cloud model, so that the probability distribution of the target point cloud data is maximized on the template point cloud model; update the transformation matrix according to the non-rigid registration mode; optimize the parameters of the Gaussian mixture model sample and the transformation matrix of the template point cloud model by repeatedly iterating the EM algorithm until the preset number of iterations is satisfied, and output the optimized transformation matrix.

[0072] In some specific embodiments, the point cloud data is loaded, and the average template and another existing ankle joint model are loaded into the same space. Further, the point cloud loaded into the same space is preprocessed. For the denoising part, radius filtering is used. A certain radius is selected, here it is 0.05, and the minimum number of points included in the neighborhood is selected as 16. The patient's ankle is downsampled, and the downsampling algorithm is used to make the number of points of the patient's ankle joint and the average template the same value, here it is 1000 points; then the CPD non-rigid point cloud registration algorithm is used to register the two point clouds. The target point cloud Y is modeled as a sample of a Gaussian mixture model (GMM), and the template point cloud X is the center point of the GMM. The entire registration process is to transform the template point cloud X into the target point cloud Y through the optimal transformation. Among them, non-rigid registration allows the point cloud to deform freely and becomes smooth through regularization constraints. Here, the template point cloud X is the average ankle joint made and marked in the previous steps, and the target point cloud Y is the new patient's ankle joint. The EM algorithm is used to gradually optimize the parameters of the GMM and the transformation matrix of the template point, and apply the transformation matrix to the template point cloud, finally making the template point cloud X closest to the target point cloud Y.

[0073] In the EM algorithm, the E-step calculation (expectation calculation) is first performed to calculate the corresponding probability matrix P between the template point cloud X and the target point cloud Y. Then, the M-step calculation (maximization calculation) is carried out to optimize the transformation parameter T of the template point cloud X, so that the probability distribution of the point cloud Y is maximized on X. According to the non-rigid registration mode, the transformation matrix T is updated. After that, the E-step calculation and the M-step calculation are repeated until the registration error converges or the set number of iterations is reached. Finally, the average template point cloud after registration is obtained. The shape of this point cloud is the same as that of the patient's ankle joint, and the marked points on it are the positions where the patient's prosthesis should be placed.

[0074] S300. Use the random forest algorithm to perform data prediction on the ankle joint image data to be processed, and obtain the ankle joint prosthesis model data.

[0075] Specifically, label and measure the ankle joint image data to be processed to obtain the key feature parameters of the ankle joint image. The key feature parameters of the ankle joint image include the longest distance of the distal tibia, the thickness of the medial malleolus of the distal tibia, the surface area of the distal tibia, and the volume of the distal tibia. Perform data cleaning on the key feature parameters of the ankle joint image to obtain the cleaned key feature parameters of the ankle joint image. Use the random forest algorithm to perform sampling training with replacement on the cleaned key feature parameters of the ankle joint image, and output the ankle joint prosthesis model data.

[0076] In this embodiment, each input key feature of the image is labeled and measured to obtain the volume size, surface area size, widest distance of the distal tibia, and thickest distance of the distal tibia of the ankle joint. Label the longest distance L of the distal tibia, the thickness D of the medial malleolus of the distal tibia, the surface area S of the distal tibia, and the volume V of the distal tibia. As Figure 13 shown, the 3D slicer tool is used in the measurement process to make the measurement results accurate, and the corresponding prosthesis models of the ankle joint are mapped one by one. There are a total of 5 models. According to the above data, a prediction model is obtained through the random forest. According to the prediction model, the prosthesis model suitable for a new ankle joint can be predicted.

[0077] In some specific embodiments, based on the data obtained for ankle joint indicators, data cleaning is required for these data, including handling missing values and outliers in the processor, and unifying the units and scales. Next, parameter initialization is performed. The following parameters need to be set: n_estimators: the number of decision trees, which is selected as 100 here. max_depth: the maximum depth of each decision tree, which is selected as 10 here. max_features: the maximum number of features that can be selected for each split, which is selected as "sqrt" here. Then, a data sampling process is carried out. Sub-samples are drawn with replacement from the original training dataset to form multiple "sub-datasets", and each sub-dataset is used to train a decision tree. Further, decision trees are constructed. For each sub-dataset, a decision tree is constructed: during the splitting process of each node, a random selection of partial features (max_features) is made. Based on the randomly selected features, the best split point is searched for. The above splitting process is repeated until one of the following conditions is met:

[0078] 1) The set maximum depth (max_depth) is reached.

[0079] 2) The number of samples in the current node is less than the set minimum number of samples for splitting (min_samples_split).

[0080] 3) The sample purity of the current node is already high enough (e.g., the Gini impurity or information gain reaches the threshold).

[0081] Finally, when training, 80% of the dataset is divided into the training set and 20% is set as the test set. As shown in Table 1, a random forest prediction model is finally obtained through training, and this model can predict the prosthesis model corresponding to the patient's ankle joint.

[0082] Table 1 Data table of the accuracy of predicting prosthesis models using a random forest

[0083] Mean Squared Error (MSE) 0.02832666 Mean Absolute Error (MAE) 0.06266666 R-squared (R2) 0.97884

[0084] S400. Combine the data of the ankle joint prosthesis placement position and the ankle joint prosthesis model to achieve three-dimensional preoperative planning for ankle joint replacement.

[0085] In summary, the improvement points of the embodiments of the present invention compared with the prior art are as follows:

[0086] 1) By using a statistical shape model, which is a modeling technique based on shape statistics and can efficiently describe shape variations. Through principal component analysis, the statistical shape model can identify the main patterns of variations and describe complex shapes with a small number of parameters. Taking the ankle joint as an example, the medial malleolus part of the ankle joint can be represented by a small number of points. The statistical shape model can also describe the differences between different individuals. Ankle joint data vary among individuals, but the model can reduce these individual differences to obtain the final average model.

[0087] 2) By establishing a statistical shape model, an average ankle joint template is obtained. Doctors only need to perform one annotation on this template, which greatly reduces the labor cost and the learning cost for doctors, providing convenience for ankle joint replacement surgery.

[0088] 3) Using a custom prosthesis positioning technique in preoperative planning for ankle joint replacement surgery. This technique starts from three-dimensional images. Three-dimensional images can provide depth information and can express the width, height, and depth of an object, while two-dimensional images only have planar information (width and height). Three-dimensional images can show the three-dimensional sense of an object and are closer to the human visual experience of the world. They support precise measurement and analysis. Three-dimensional images provide precise spatial coordinates and can be used to measure geometric properties such as distance, volume, and angle. Through three-dimensional images, multi-dimensional data analysis such as surface curvature, area, and volume can be carried out. Most importantly, three-dimensional images have stronger analysis capabilities compared to two-dimensional images. Three-dimensional images can more precisely describe and simulate the physical processes in the real world through mathematical modeling. Three-dimensional images can capture shape and spatial information that two-dimensional images cannot capture, facilitating classification, detection, and reconstruction. Compared with the original surgical planning that annotates on two-dimensional X-rays, the technique for surgical planning in three-dimensional space is more accurate and faster, and greatly reduces the learning cost and hospital expenses.

[0089] 4) Using a custom registration technique, image preprocessing is added to the CPD algorithm. Through the previous preprocessing, the patient's ankle joint model and the average template are made lightweight, making subsequent processing fast and simple. The CPD algorithm has higher flexibility compared to other algorithms and can be applied to both rigid and non-rigid registration tasks. It can capture complex non-linear deformations during application and is very suitable for the registration task in preoperative planning. Moreover, during registration, no additional manual marking is required, reducing human interference and ensuring the authenticity of the results. Since no additional manual operation is needed, this method also appears more intelligent. Compared with other algorithms, this algorithm has higher registration accuracy.

[0090] 5) The method for determining the prosthesis model in the customized preoperative planning of ankle joint replacement uses the method of machine learning in mathematical modeling. After establishing the data set, it is trained by using the random forest to read the data. The random forest belongs to a model with high accuracy in the prediction model. By combining the prediction results of each decision tree, it can obtain a higher prediction accuracy than a single decision tree. And its sampling process is random sampling, which reduces the overfitting phenomenon of a single decision tree. In the process of use, the random forest model is less sensitive to parameters, and the main parameters are easy to understand and adjust, which is convenient for the use of learners and has higher convenience compared with other models.

[0091] Referring to Figure 2 , a preoperative planning system based on three-dimensional CT images of ankle joint replacement, comprising:

[0092] The first module 201 is used to process the ankle joint image data to be processed through a statistical model and construct a three-dimensional average ankle joint point cloud model;

[0093] The second module 202 is used to register the three-dimensional average ankle joint point cloud model with the ankle joint image data to be processed through the CPD non-rigid point cloud registration algorithm to obtain the placement position of the ankle joint prosthesis;

[0094] The third module 203 is used to predict the ankle joint image data to be processed through the random forest algorithm to obtain the ankle joint prosthesis model data;

[0095] The fourth module 204 is used to combine the placement position of the ankle joint prosthesis with the ankle joint prosthesis model data to realize the preoperative planning of three-dimensional ankle joint replacement.

[0096] The content in the above method embodiments is applicable to the system embodiments. The functions specifically implemented by the system embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those of the above method embodiments.

[0097] The above is a specific description of the preferred embodiments of the present invention. However, the present invention is not limited to the described embodiments. Those skilled in the art can make various equivalent deformations or substitutions without departing from the spirit of the present invention, and these equivalent deformations or substitutions are all included in the scope defined by the claims of this application.

Claims

1. A preoperative planning method for ankle replacement based on three-dimensional CT images, characterized in that: The following steps are involved: The ankle joint image data to be processed is processed by using a statistical model to construct a three-dimensional average ankle joint point cloud model; The three-dimensional average ankle joint point cloud model is registered with the ankle joint image data to be processed by the CPD non-rigid point cloud registration algorithm to obtain the placement position of the ankle joint prosthesis; The ankle joint image data to be processed is predicted by using a random forest algorithm to obtain ankle joint prosthesis model data; Combining the ankle prosthesis placement position and ankle prosthesis model data, three-dimensional ankle replacement preoperative planning can be achieved.

2. A preoperative planning method for ankle replacement based on three-dimensional CT images according to claim 1, characterized in that: The step of processing the ankle joint image data to be processed by using a statistical model to construct a three-dimensional average ankle joint point cloud model specifically includes: Acquire ankle joint image data to be processed; Modeling and processing the ankle joint image data to be processed by using a statistical model to obtain a three-dimensional average ankle joint model; Performing data annotation processing on the three-dimensional average ankle joint model to obtain a three-dimensional average ankle joint model with annotations; The annotated three-dimensional average ankle joint model is transformed through the average grid sampling algorithm to obtain a three-dimensional average ankle joint point cloud model.

3. A preoperative planning method for ankle replacement based on three-dimensional CT images according to claim 2, characterized in that: The step of modeling the ankle joint image data to be processed by using a statistical model to obtain a three-dimensional average ankle joint model specifically includes: Extract data from the ankle joint image data to be processed, and obtain a number of distal tibia sample data; Performing rigid alignment on a number of distal tibia sample data to obtain aligned distal tibia sample data; Performing data normalization processing on the aligned distal tibia sample data to obtain normalized distal tibia sample data; Extract feature points from the normalized distal tibia sample data to obtain distal tibia sample feature data; Perform principal component analysis on the characteristic data of distal tibia samples to obtain the principal component eigenvector; The shape of the tibia is represented as the mean shape and combined with the principal component eigenvectors to construct a three-dimensional average ankle joint model.

4. A preoperative planning method for ankle replacement based on three-dimensional CT images according to claim 3, characterized in that: The step of registering the three-dimensional average ankle joint point cloud model with the ankle joint image data to be processed by the CPD non-rigid point cloud registration algorithm to obtain the ankle joint prosthesis placement position specifically includes: The ankle joint image data to be processed is converted by using an average grid sampling algorithm to obtain ankle joint three-dimensional point cloud image data; Loading the ankle joint three-dimensional point cloud image data and the three-dimensional average ankle joint point cloud model into the same space and performing radius filtering processing to obtain target point cloud data and a template point cloud model, wherein the template point cloud model is derived from the three-dimensional average ankle joint point cloud model, and the target point cloud data is derived from the ankle joint three-dimensional point cloud image data; The target point cloud data and the template point cloud model are registered using the CPD non-rigid point cloud registration algorithm to obtain the placement position of the ankle prosthesis.

5. A preoperative planning method for ankle replacement based on three-dimensional CT images according to claim 4, characterized in that: The step of registering the target point cloud data with the template point cloud model by using the CPD non-rigid point cloud registration algorithm to obtain the ankle prosthesis placement position specifically includes: Based on the target point cloud data, a Gaussian mixture model sample is constructed; The parameters of the Gaussian mixture model sample and the transformation matrix of the template point cloud model are optimized through the EM algorithm, and the optimized transformation matrix is ​​output; The optimized transformation matrix is ​​applied to the template point cloud model for registration to obtain the placement position of the ankle prosthesis.

6. The preoperative planning method for ankle replacement based on three-dimensional CT images according to claim 5, characterized in that: The step of optimizing the parameters of the Gaussian mixture model sample and the transformation matrix of the template point cloud model by the EM algorithm and outputting the optimized transformation matrix specifically includes: Perform expectation calculation on the Gaussian mixture model samples to obtain the probability distribution between the target point cloud data and the template point cloud model; Maximize the calculation of the Gaussian mixture model samples and optimize the transformation matrix of the template point cloud model to maximize the probability distribution of the target point cloud data on the template point cloud model. According to the non-rigid registration mode, the transformation matrix is ​​updated; Repeat the step of optimizing the parameters of the Gaussian mixture model samples and the transformation matrix of the template point cloud model through the EM algorithm until the preset number of iterations is met and the optimized transformation matrix is ​​output.

7. A preoperative planning method for ankle replacement based on three-dimensional CT images according to claim 6, characterized in that: The step of performing data prediction on the ankle joint image data to be processed by using a random forest algorithm to obtain ankle joint prosthesis model data specifically includes: Annotate and measure the ankle joint image data to be processed to obtain key characteristic parameters of the ankle joint image, wherein the key characteristic parameters of the ankle joint image include the longest distance of the distal tibia, the thickness of the medial malleolus of the distal tibia, the surface area of ​​the distal tibia, and the volume of the distal tibia; Performing data cleaning on key feature parameters of the ankle joint image to obtain cleaned key feature parameters of the ankle joint image; The key feature parameters of the cleaned ankle joint images are extracted with replacement through the random forest algorithm to output the ankle prosthesis model data.

8. A preoperative planning system for ankle replacement based on three-dimensional CT images, characterized in that: Includes the following modules: The first module is used to process the ankle joint image data to be processed through a statistical model to construct a three-dimensional average ankle joint point cloud model; The second module is used to register the three-dimensional average ankle joint point cloud model with the ankle joint image data to be processed through the CPD non-rigid point cloud registration algorithm to obtain the placement position of the ankle joint prosthesis; The third module is used to predict the ankle joint image data to be processed by using a random forest algorithm to obtain ankle joint prosthesis model data; The fourth module is used to combine the ankle prosthesis placement position and ankle prosthesis model data to achieve three-dimensional ankle replacement preoperative planning.

Images