WO2023142956A1 - Total hip replacement preoperative planning system based on deep learning - Google Patents

Abstract

The present application provides a total hip replacement preoperative planning system based on deep learning, comprising: a total hip joint image acquisition module which is used for obtaining a total hip joint three-dimensional block diagram to be identified; a total hip joint identification module which is used for inputting the total hip joint three-dimensional block diagram to be identified into a trained three-dimensional segmentation neural network to obtain a femur area in each total hip joint two-dimensional cross-sectional image, wherein the trained three-dimensional segmentation neural network is obtained by training a convolutional neural network by means of preset three-dimensional block diagrams labeled with labels of femur areas; and a total hip joint three-dimensional image construction module which is used for, according to the femur area in each total hip joint two-dimensional cross-sectional image and on the basis of three-dimensional reconstruction technology, obtaining a three-dimensional image of the femur area. According to the present application, total hip joint three-dimensional block diagrams are identified, three-dimensional modeling is performed on extracted femur areas on the basis of the three-dimensional block diagrams, and thus, the total hip joint identification precision is improved according to the femur area three-dimensional model.

Description

Preoperative planning system for total hip replacement based on deep learning

Cross References to Related Applications

This application claims the priority of the Chinese patent application with application number 202210101412.9 and titled "Deep Learning-Based Preoperative Planning System for Total Hip Joint Replacement" filed on January 27, 2022, which is incorporated herein by reference in its entirety.

technical field

This application relates to the technical field of artificial intelligence, in particular to a preoperative planning system for total hip replacement based on deep learning.

Background technique

Joint replacement refers to the use of materials such as metal, polymer polyethylene or ceramics to make artificial joint prostheses according to the shape, structure and function of human joints, and implant them into the human body through surgical techniques.

In the preoperative planning of total hip replacement, physicians need to rely on their own experience to judge the entire femoral region from medical images of the hip joint, and determine the size and type of the prosthesis that needs to be replaced. However, traditional total hip replacement technology relies on physician experience, and physicians with different experience will have different identification results, making it difficult to guarantee the uniformity of results. In recent years, with the improvement of medical level, the use of two-dimensional image segmentation neural network recognition can eliminate this shortcoming.

Since the shape of the hip joint is a three-dimensional structure, using a two-dimensional image segmentation neural network for total hip joint segmentation will lose the feature information between the continuous slice layers of the joint, resulting in low recognition accuracy of the total hip joint.

Contents of the invention

In view of the problems existing in the prior art, the present application provides a preoperative planning system for total hip replacement based on deep learning.

The present application provides a preoperative planning system for total hip joint replacement based on deep learning, including: a total hip joint image acquisition module, which is used to acquire a three-dimensional block diagram of the total hip joint to be identified, and the three-dimensional block image to be identified The picture is formed by stacking multiple two-dimensional cross-sectional images of the total hip joint; the total hip joint identification module is used to input the three-dimensional block diagram of the total hip joint to be identified into the trained three-dimensional segmentation neural network, Obtaining the femoral region in each two-dimensional cross-sectional image of the total hip joint output by the trained three-dimensional segmentation neural network, the trained three-dimensional segmentation neural network of the total hip joint is preset by a label marked with the femoral region The 3D block diagram of the total hip joint is used as a training sample, which is obtained by training the convolutional neural network; the 3D image construction module of the total hip joint is used for 3D reconstruction based on the femur in each 2D cross-sectional image of the total hip joint. region to obtain a three-dimensional image of the femoral region.

According to a deep learning-based preoperative planning system for total hip replacement provided by the present application, the total hip recognition module is also used to: acquire several two-dimensional cross-sectional images of sample hip joints; Annotate the sample femoral region in the two-dimensional cross-sectional image of the joint, and mark the femoral head region label on the femoral head pixels in the sample femoral region to obtain several first preset two-dimensional cross-sectional images of the hip joint; The acquisition sequence of the two-dimensional cross-section is to stack several first preset two-dimensional cross-sectional images of the hip joint to obtain the corresponding first preset three-dimensional block image; The shape image is input to the initial three-dimensional segmentation neural network for training to obtain a trained three-dimensional segmentation neural network; wherein, the initial three-dimensional segmentation neural network is constructed by the U-Net convolutional network, and the U The convolution kernel of the -Net convolutional network is a three-dimensional convolution kernel.

According to a deep learning-based preoperative planning system for total hip joint replacement provided by the present application, the total hip joint recognition module, when inputting a plurality of the first preset three-dimensional block images into the initial The three-dimensional segmentation neural network is trained, and after the trained three-dimensional segmentation neural network is obtained, it is also used to: mark the cortical bone region labels on the cortical bone pixels of the sample femur region in the first preset two-dimensional cross-sectional images of the hip joint , obtaining several second preset two-dimensional cross-sectional images of the hip joint; stacking the several second preset two-dimensional cross-sectional images of the hip joint in order to obtain a corresponding second preset three-dimensional block diagram; A plurality of the second preset three-dimensional block diagrams are used to optimize the parameters of the trained three-dimensional segmentation neural network to obtain a hip joint recognition model.

According to a deep learning-based preoperative planning system for total hip replacement provided by the present application, the total hip recognition module, after acquiring the three-dimensional block diagram of the total hip to be recognized, is further used to: if The three-dimensional block diagram of the total hip joint to be identified is a three-dimensional block diagram of the hip joint, and the three-dimensional block diagram of the hip joint is input into the hip joint identification model to obtain the femur in each two-dimensional cross-sectional image of the hip joint. area and cortical bone area; according to the femoral area and the cortical bone area, determine the medullary canal area; calculate the midpoint coordinates of each medullary canal level in the medullary canal area, and according to the midpoint coordinates, all Carry out straight line fitting at the center point to determine the anatomical axis of the medullary cavity; calculate the angle value of the femoral neck-shaft angle according to the anatomical axis of the medullary cavity and the axis of the femoral neck; Center position, to determine the type and placement of the femoral stem prosthesis model.

According to a preoperative planning system for total hip joint replacement based on deep learning provided by the present application, the three-dimensional image construction module of the total hip joint, in the three-dimensional reconstruction technology based on the three-dimensional reconstruction technology, according to each two-dimensional cross-sectional image of the total hip joint After obtaining the three-dimensional image of the femoral region, it is also used to: obtain the pixel coordinates of the femoral head center point of the femoral region in the three-dimensional image based on the centroid formula of the planar image; convert the pixel coordinates into an image coordinates; determine the position of the center of rotation of the femoral head; obtain first size information according to the position of the center of rotation of the femoral head, to determine second size information based on the first size information, wherein the first size information is the size corresponding to the femoral head Information, the second size information is the size information corresponding to the acetabular cup prosthesis model.

According to a deep learning-based preoperative planning system for total hip arthroplasty provided by the present application, the preoperative planning system further includes: a position correction module, which is used to place the placed femoral stem prosthesis model The position or the placement position of the acetabular cup prosthesis model is corrected, so that the placement position of the femoral stem prosthesis model and the placement position of the acetabular cup prosthesis model meet the preset position requirements.

According to a preoperative planning system for total hip arthroplasty based on deep learning provided by the present application, the three-dimensional image construction module of the total hip joint, when acquiring the first size information according to the position of the femoral head rotation center, is specifically used to: determine The intersection point between the rotation center position of the femoral head and the edge of the femoral head area; the first size information is obtained through the length value between the intersection point and the center position of the femoral head.

According to a preoperative planning system for total hip arthroplasty based on deep learning provided by the present application, the three-dimensional image construction module of the total hip joint determines the intersection point between the center of rotation of the femoral head and the edge of the femoral head area, and passes the intersection point and the center of the femoral head The length value between the positions, when obtaining the first size information, is specifically used to determine multiple intersections between the center of rotation of the femoral head and the edge of the femoral head area at different angles; through the multiple intersections and the central position of the femoral head The average or median of multiple length values to obtain the first size information.

According to a deep learning-based preoperative planning system for total hip replacement provided by the present application, two continuous three-dimensional convolution kernels in the convolutional neural network are connected through a residual structure, and the convolutional neural network The loss function of is composed of DICE loss and BCE loss.

According to a preoperative planning system for total hip arthroplasty based on deep learning provided by the present application, the total hip joint recognition module is specifically used to determine the medullary cavity area according to the femoral area and the cortical bone area. : filter out the cortical bone area in the femoral area to obtain the medullary cavity area.

The above technical solution of the present application has at least the following beneficial effects:

The preoperative planning system for total hip replacement based on deep learning provided by this application stacks multiple two-dimensional cross-sectional images of the total hip joint into a three-dimensional block diagram of the total hip joint, and is based on a convolutional neural network with a three-dimensional convolution kernel structure. Identify the three-dimensional block diagram of the total hip joint, effectively extract the femoral region of each two-dimensional cross-sectional image of the total hip joint, and perform three-dimensional modeling based on the extracted femoral region to obtain a more accurate three-dimensional model of the femoral region , so as to improve the recognition accuracy of the total hip joint during preoperative planning of total hip joint replacement according to the three-dimensional model of the femoral region.

Description of drawings

In order to more clearly illustrate the technical solutions in this application or the prior art, the accompanying drawings that need to be used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the accompanying drawings in the following description are the present For some embodiments of the application, those skilled in the art can also obtain other drawings based on these drawings without creative work.

Fig. 1 is a schematic flow chart of the preoperative planning method for total hip arthroplasty based on deep learning provided by the present application;

Fig. 2 is a schematic structural diagram of the three-dimensional residual U-Net convolutional neural network provided by the present application;

Fig. 3 is a schematic diagram of the 3D convolution process provided by the present application;

Fig. 4 is the femur region identification rendering based on 3D Res U-Net provided by the application;

Fig. 5 is the schematic diagram of a kind of acetabular cup prosthesis model plan provided by the present application;

Fig. 6 is a schematic diagram of a femoral stem prosthesis model plan provided by the present application;

Fig. 7 is the effect diagram of the bone cortex region recognition based on 3D Res U-Net provided by the present application;

FIG. 8 is a schematic flowchart of another deep learning-based preoperative planning method for total hip arthroplasty provided by the present application;

FIG. 9 is a schematic structural diagram of a preoperative planning system for total hip arthroplasty based on deep learning provided by the present application;

FIG. 10 is a schematic structural diagram of an electronic device provided by the present application.

Detailed ways

In order to make the purpose, technical solutions and advantages of this application clearer, the technical solutions in this application will be clearly and completely described below in conjunction with the accompanying drawings in this application. Obviously, the described embodiments are part of the embodiments of this application , but not all examples. Based on the embodiments in this application, all other embodiments obtained by persons of ordinary skill in the art without creative efforts fall within the protection scope of this application.

Figure 1 is a schematic flowchart of the preoperative planning method for total hip arthroplasty based on deep learning provided by this application. As shown in Figure 1, this application provides a preoperative planning method for total hip arthroplasty based on deep learning, including:

Step 101 , acquiring a three-dimensional block diagram of the total hip joint to be identified, where the three-dimensional block diagram of the total hip joint to be identified is formed by stacking multiple two-dimensional cross-sectional images of the total hip joint.

In this application, firstly, based on the digital imaging and communications in medicine (DICOM) data of the preoperative planning of the total hip joint, the medical image dataset of the total hip joint to be identified is constructed. The image is a two-dimensional cross-sectional image of the total hip joint. Different two-dimensional cross-sectional images of the total hip joint correspond to different bone regions. The bone area corresponding to the image is the tibia area (including a part of the femur area). Image of the right femur; then, convert the 2D cross-sectional DICOM data of the hip joint into a JPG image, and convert multiple converted hip joint The cross-sectional images are stacked to form a three-dimensional block map of the total hip joint to be identified, and the three-dimensional block map of the hip joint is obtained.

Step 102, input the three-dimensional block diagram of the total hip joint to be identified into the trained three-dimensional segmentation neural network, and obtain each two-dimensional cross-sectional image of the total hip joint output by the trained three-dimensional segmentation neural network The femoral region, the trained three-dimensional segmentation neural network is obtained by training the convolutional neural network from a preset three-dimensional block diagram marked with a label of the femoral region.

In this application, the stacked 3D block diagram of the hip joint is recognized through the trained 3D segmentation neural network. In this application, the convolution kernel of the convolutional neural network used to construct the three-dimensional segmentation neural network is a three-dimensional convolution kernel, so that the three-dimensional segmentation neural network obtained based on the convolutional neural network training can be used for image segmentation. Feature extraction is carried out in three dimensions of the three-dimensional block map of the hip joint, effectively extracting the feature information of the femoral region between each cross-section, and finally segmenting the three-dimensional block map corresponding to the femoral region. During the training process of the three-dimensional segmentation neural network, the sample femur region of the three-dimensional block image of the sample hip joint has been marked, so that the three-dimensional segmentation neural network can directly identify the femur region on the three-dimensional hip joint image. It should be noted that, when the application is training the convolutional neural network, the pixels of the femoral head in the sample femoral area in the three-dimensional block diagram of the sample hip joint are also marked, so that the model can identify the entire femoral area while also being able to The femoral head region is further identified, so that the femoral region in the hip joint can be quickly identified, saving time and cost, and improving the recognition accuracy of the hip joint. It should be noted that if it is necessary to perform image recognition on the two-dimensional cross-sectional image of the knee joint, in addition to obtaining the two-dimensional cross-sectional image of the hip joint provided in the above embodiment, it also includes a two-dimensional cross-sectional image of the knee joint (which can be reconstructed through three-dimensional, Construct the 3D images of the tibia region and the patella region), and then segment the femur region through the trained 3D segmentation neural network, and then obtain the end of the femur region, and the 3D images of the tibia and patella region according to the segmentation recognition, and determine the relationship between the knee joint and the tibia Prosthesis type and installation position.

Step 103, based on the three-dimensional reconstruction technology, according to the femoral region in each two-dimensional cross-sectional image of the total hip joint, obtain the three-dimensional image of the femoral region.

In this application, the identified femoral region is formed by stacking multiple two-dimensional cross-sectional images, and then three-dimensional information is reconstructed by using key point information in multiple two-dimensional images through three-dimensional reconstruction technology, thereby obtaining a three-dimensional image of the femur.

The preoperative planning method for total hip arthroplasty based on deep learning provided by this application stacks multiple two-dimensional cross-sectional images of the total hip joint into a three-dimensional block diagram of the total hip joint, and based on the convolutional neural network of the three-dimensional convolution kernel structure, Identify the three-dimensional block diagram of the total hip joint, effectively extract the femoral region of each two-dimensional cross-sectional image of the total hip joint, and perform three-dimensional modeling based on the extracted femoral region to obtain a more accurate three-dimensional model of the femoral region , so as to improve the recognition result of the total hip joint during the preoperative planning of the total hip joint replacement according to the three-dimensional model of the femoral region.

On the basis of the foregoing embodiments, the trained three-dimensional segmentation neural network is obtained through the following steps:

The preset three-dimensional block image set is used as a training sample, input to the initial three-dimensional segmentation neural network for training, and a trained three-dimensional segmentation neural network is obtained;

Wherein, the set of preset three-dimensional block images includes several preset three-dimensional block images obtained by stacking preset two-dimensional cross-sectional images.

In this application, different medullary cavities in the preset three-dimensional block diagram correspond to different bone regions, and the corresponding preset three-dimensional block diagram is constructed according to the actual bone region segmentation requirements. For example, for the preoperative planning of the femoral head and femoral stem prosthesis, the preset 3D block diagram is formed by stacking the 2D cross-sectional images of the hip joint; if the preoperative planning of the knee joint and tibial prosthesis is required, the preset In addition to the 2D cross-sectional image of the hip joint, the 3D block diagram also includes the corresponding 2D cross-sectional images of the patella region and the tibial region. After identifying the femur region through the trained 3D segmentation neural network, it is combined with the existing 3D reconstruction Technology, based on the two-dimensional cross-sectional images of the patella region and the tibia region, constructs a three-dimensional image of the patella and tibia, thereby providing more accurate image data for the preoperative planning of the knee joint and tibial prosthesis.

On the basis of the foregoing embodiments, the training process specifically includes:

Step 201, acquiring several two-dimensional cross-sectional images of the sample hip joint;

Step 202, labeling the sample femoral region in each sample hip joint two-dimensional cross-sectional image, and marking the femoral head region label on the femoral head pixels of the sample femoral region, and obtaining several first preset two-dimensional hip joint images. cross-sectional images;

Step 203: According to the acquisition sequence of the sample hip joint two-dimensional cross-section, stack several first preset two-dimensional cross-sectional images of the hip joint to obtain a corresponding first preset three-dimensional block image;

Step 204, inputting a plurality of the first preset three-dimensional block images to the initial three-dimensional segmentation neural network for training to obtain a trained three-dimensional segmentation neural network;

Wherein, the initial three-dimensional segmentation neural network is constructed by the U-Net convolution network, and the convolution kernel of the U-Net convolution network is a three-dimensional convolution kernel.

In this application, the hip joint medical image sample data set is obtained, the sample hip joint two-dimensional cross-sectional image in the sample set is manually labeled with the sample femur region, and only the label containing the femur part is extracted as a mask mask. Since this application can also extract the femoral head part of the femoral region while extracting the femoral region, the pixels of the femoral head in the sample femoral region are also marked during labeling, and finally the marked sample hip joint two The three-dimensional cross-sectional images (that is, the first preset two-dimensional cross-sectional image of the hip joint) are stacked to obtain the first preset three-dimensional block image. Specifically, this application converts the DICOM data of the first preset two-dimensional cross-section of the hip joint into a picture in JPG format, and at the same time, converts the annotation file into a picture in PNG format. The ratio is divided into training set, validation set and test set. Since the input image of the convolutional neural network of the present application has one more dimension than the input image of the existing 2D network, that is, the three-dimensional block diagram of the sample hip joint (the first preset three-dimensional block image) is formed by stacking multiple two-dimensional cross-sectional images Therefore, the annotation file corresponds to the 3D block diagram of the sample hip joint, which is also a block diagram. It should be noted that, in this application, the first preset three-dimensional block image is obtained by pre-extracting and stacking two-dimensional cross-sections of the femoral part, and the femoral head region is calculated on the three-dimensional block image of the femoral part stack. Labeling enables the segmentation network to more quickly segment the femoral head region in the image from the background after the training is completed.

Fig. 2 is a schematic structural diagram of the three-dimensional residual U-Net convolutional neural network provided by the application, as shown in Fig. 2, in this application, the three-dimensional residual U-Net convolutional neural network (referred to as 3D Res UNet ) is built based on U-Net, and also includes an Encoder (encoder) part and a Decoder (decoder) part, where the Encoder part is used to analyze the entire picture and perform feature extraction and analysis, and with it The corresponding Decoder part is the process of restoring features. By training the convolutional neural network, the obtained 3D segmentation neural network can segment the femur region in the 3D block image to be recognized, and obtain the segmented femoral region block state diagram.

Specifically, the Encoder part is composed of the basic residual module ResBlock and the maximum pooling layer MaxPooling. Its input size (Input shape) is 1*8*256*256*1. ResBlock is composed of basic convolutional blocks, including two A continuous 3D convolution kernel, each 3D convolution kernel includes two sets of operations: 2*(conv+relu+bn), that is, 3D convolution (conv), activation function (relu) and batch normalization (Batch Normalization , referred to as BN) constitute a set of operations, among which, 3D convolution can effectively extract the information between the cross-sections and reduce the false detection rate; the activation function can increase the nonlinear ability of the model and improve the feature extraction ability of the model; batch normalization The optimization operation can change the distribution of data, which is conducive to the rapid convergence of network training. Figure 3 is a schematic diagram of the 3D convolution process provided by this application, as shown in Figure 3, since the convolution kernel is three-dimensional, it can slide in a certain step in three dimensions, and it will surround 3*3*3 The information of the region is combined to form a point, which can extract richer feature information.

Preferably, two consecutive three-dimensional convolution kernels in the convolutional neural network are connected through a residual structure. As shown in Figure 2, in this application, two consecutive 3D convolution kernels can prevent network degradation through a residual connection (Skip Connection). Connect the two 3D convolution kernels through the residual structure, and add the feature maps in the network before and after the continuous convolution blocks, so that the network can choose the appropriate backpropagation during training. path. In addition, the Maxpooling operation uses the maximum value of the adjacent fixed-size region as the characteristic representation of the region, which can effectively reduce the calculation of network parameters.

Further, the Decoder part is composed of upsampling and convolution blocks (deconvolution Deconv). In the Decoder layer, the Encoder feature map and the upsampled feature map will be channel-stacked (Concat), and then deconvolved through the convolution block of the Decoder part, and its output shape (output shape) is 1*8* 256*256*1, where the composition of the convolution block is the same as the Encoder process, and will not be repeated here.

In this application, when 3D Res UNet is training, the batch size of samples fed to the network each time is 8, the initial learning rate is set to 1e-4, and every 5000 iterations (iterations), the learning rate decays to the original 0.95 . Preferably, in the present application, the optimizer uses an Adam optimizer, and the loss function of the convolutional neural network is composed of DICE loss and BCE loss (DICE loss and BCE loss are respectively a type of loss function). Since the loss function used is a fusion of DICE loss and BCE loss, this can avoid oscillations during network training when only DICE loss is used.

Further, based on the training set, verification set and test set divided in the previous period, set each iteration 1000 times, do a verification on the training set and verification set, and measure the model's train loss (training error) and val loss (verification error) , as well as train DICE and val DICE, use the early stopping method to judge the timing of network training stop, and get the trained three-dimensional segmentation neural network. In addition, in the testing phase, the DICOM data and annotation files of the two-dimensional cross-section of the hip joint of the entire sample case were converted into images in JPG and PNG formats in sequence, and packaged into image blocks (ie, block diagrams), which passed the test. Get test DICE.

On the basis of the above-mentioned embodiments, after obtaining the three-dimensional image of the femoral region according to the femoral region in each two-dimensional cross-sectional image of the total hip joint based on the three-dimensional reconstruction technology, the method further includes:

Based on the centroid formula of the planar image, obtain the pixel coordinates of the center point of the femoral head in the femoral region in the three-dimensional image;

converting said pixel coordinates to image coordinates;

Determine the position of the center of rotation of the femoral head;

According to the rotation center position of the femoral head, the first size information is obtained to determine the second size information according to the first size, wherein the first size information is the size information corresponding to the femoral head, and the second size information is the hip The size information corresponding to the acetabular cup prosthesis model. Optionally, in the present application, the size information includes at least a diameter and a radius, for example, by obtaining the diameter of the femoral head, the diameter of the acetabular cup prosthesis model is determined.

In this application, 3D Res U-Net is used to identify each pixel area of the three-dimensional block map of the hip joint. During the training process, the pixel label can be divided into two attribute values, named 0 and 1 respectively, where the value 0 Represents the background pixel, 1 represents the femoral head pixel; after the labeling is completed, the marked image data is passed into the convolutional neural network (ie 3D Res UNet), and the convolution pooling sampling is used to iteratively learn and train. Fig. 4 is the effect diagram of the femoral region recognition based on 3D Res U-Net provided by the present application. As shown in Fig. 4, after the training is completed, the trained three-dimensional segmentation neural network can automatically recognize the position of the femoral head and complete the femoral head region identification Identification (marked schematically with a white line in Figure 4). Finally, through 3D reconstruction, a 3D image of the femoral head region is obtained.

After obtaining the 3D image of the femoral head region, since the image output by the 3D segmentation neural network is a binary image, in the binary image, there are only two pixel values of "0" and "1", and its mass distribution is uniform , so in the recognized 3D image of the femoral region, the centroid and centroid coincide. According to the centroid formula of the planar image, the coordinates of the center point of the femoral head in the 3D image of the femoral region can be obtained, that is, the femoral head rotation in the 3D image center. Specifically, if the binary image is B, then B[i, j] represents the pixel value of the i-th row and j-th column pixel in the binary image B, so the following formula can be used to obtain the femoral head of the femoral region in the three-dimensional image The location of the center point:

Among them, A represents the sum of the pixel values of all pixels in the binary image,

n represents the maximum number of rows of pixels in the binary image, and m represents the maximum number of columns of pixels in the binary image, so that the pixel coordinates of the center point of the femoral head can be obtained, and then the pixel coordinates are converted into image coordinates.

Specifically, the center coordinates of the image plane coordinates are:

Then the pixel coordinates

The transformation formula to image coordinates (x', y') is:

Wherein, S _x , S _y are the row and column spacing of the image array respectively. Finally, according to the image coordinates, the position of the center of rotation of the femoral head in the three-dimensional image is obtained, and the diameter of the femoral head is determined. Since the diameter of the femoral head is equal to the diameter of the inner circle of the acetabular cup, the diameter of the acetabular cup prosthesis model is calculated. Exemplarily, the diameter of the femoral head is based on the center of the femoral head, intersects a ray with the edge of the femoral head region obtained based on the preoperative planning method for total hip arthroplasty provided by this application, calculates the length from the intersection point to the center of the femoral head, and then calculates the length from the intersection point to the center of the femoral head One degree is the step length, and it is calculated once per one degree of rotation. Finally, the radius is obtained through the statistical mean value, so as to obtain the diameter of the acetabular cup prosthesis model, which ensures that the specification of the acetabular cup prosthesis model is determined according to the recognition results of the model (for example, Figure 5 provides a schematic diagram of the model plan of the acetabular cup prosthesis).

On the basis of the above embodiments, after inputting a plurality of the first preset three-dimensional block images into the initial three-dimensional segmentation neural network for training and obtaining a trained three-dimensional segmentation neural network, the method Also includes:

Marking the cortical bone region labels on the cortical bone pixels of the sample femur region in the first preset two-dimensional cross-sectional images of the hip joint to obtain several second preset two-dimensional cross-sectional images of the hip joint;

Stacking several second preset hip joint two-dimensional cross-sectional images in order to construct a corresponding second preset three-dimensional block diagram;

Through a plurality of the second preset three-dimensional block diagrams, the trained three-dimensional segmentation neural network is fine-tuned to obtain a hip joint recognition model.

In this application, by marking a new label in the first preset two-dimensional cross-sectional image of the hip joint, the parameters of the three-dimensional segmentation neural network are optimized according to the obtained new training set, so that the model recognizes the femoral head region while , but also to identify cortical areas.

On the basis of the above embodiments, after the acquisition of the three-dimensional block diagram of the total hip joint to be identified, the method further includes:

Inputting the three-dimensional block diagram of the hip joint into the hip joint recognition model to obtain the femoral region and the cortical bone region in each two-dimensional cross-sectional image of the hip joint;

determining the medullary cavity area according to the femoral area and the cortical bone area;

calculating the midpoint coordinates of each medullary canal level in the medullary cavity region, and performing straight line fitting on all the midpoints according to the midpoint coordinates to determine the anatomical axis of the medullary cavity;

Calculate the angle value of the femoral neck-shaft angle according to the anatomical axis of the medullary cavity and the femoral neck axis;

According to the angle value, the medullary cavity area and the position of the center of rotation of the femoral head, the type and placement position of the femoral stem prosthesis model are determined (for example, FIG. 6 provides a schematic diagram of a femoral stem prosthesis model plan).

In this application, the three-dimensional segmentation neural network after parameter optimization, that is, the hip joint recognition model, when recognizing each pixel region of the image, the pixel label is divided into three attribute values, named 0, 1 and 2 respectively, where the value 0 represents background pixels, 1 represents femoral head pixels, and 2 represents cortical bone. Figure 7 is the effect diagram of the recognition of the cortical bone region based on 3D Res U-Net provided by the present application. As shown in Figure 7, after optimizing the parameters of the three-dimensional segmentation neural network, the femoral head in the block diagram can be simultaneously identified. and identification of cortical bone (schematically marked with a black line in Figure 7).

Further, the present application intercepts images from the end of the lesser trochanter to the end of the femur from the recognition results output by the hip joint recognition model, and subtracts the cortical bone region from the femoral region in the image to obtain the medullary cavity region. Then, below the end position of the lesser trochanter, each row (i.e., the level of the medullary canal is identified from the two-dimensional cross-section of each hip joint, and each row refers to the two-dimensional cross-sectional image of the hip joint segmented by this application, after three-dimensional Reconstruct the simulated X-ray projection effect map, and then start below the end position of the lesser trochanter, draw a horizontal line on the image at certain preset positions, and when the horizontal line intersects with the edges of the two femoral medullary canals, four points will be obtained) and The intersection of the medullary cavity is four coordinates, which are named A1, A2, B1, and B2 from left to right; the midpoint can be obtained based on two points, namely A1 (X ₁ , Y ₁ ), A2 (X ₂ , Y ₂ ) The midpoint coordinates of can be obtained by the following formula:

B1 and B2 can be calculated in the same way. The coordinates of the midpoint of the medullary cavity are calculated in turn for each row, and these points are fitted into a straight line to determine the anatomical axis of the medullary cavity. Finally, according to the anatomical axis of the medullary cavity and the axis of the femoral neck, the angle value of the femoral neck-shaft angle was calculated, combined with the shape of the medullary cavity and the position of the center of rotation of the femoral head, the model and placement position of the femoral stem prosthesis model could be determined together. Specifically, the femoral neck-shaft angle is the angle between the anatomical axis of the medullary canal and the femoral neck axis. With these obtained parameters as the screening conditions, the existing prosthesis library is filtered, and the corresponding The optimal model; further, by moving the femoral stem prosthesis, the rotation center position of the femoral stem prosthesis coincides with the previously calculated acetabular cup rotation center position (that is, the femoral head rotation center position), and the femoral stem prosthesis is obtained. The actual location of the body. The application obtains the femoral region based on convolutional neural network identification, which can quickly and accurately determine the specification, model and placement position of the prosthesis model.

After determining the second size information of the acetabular cup prosthesis and the type and position of the femoral stem prosthesis, simulate the placement of the acetabular cup prosthesis and the femoral stem prosthesis into the target position (as shown in Figure 5 and Figure 6) . If the placement position of the acetabular cup prosthesis and the placement position of the femoral stem prosthesis meet the preset position requirements, the preoperative planning plan for the total hip joint can be output. Among them, the preset position requirements of the acetabular cup prosthesis can be, for example, that after the acetabular cup prosthesis is placed in the acetabular socket, the coverage rate covering the acetabular socket is greater than 75%, and the preset position requirements of the femoral stem prosthesis can be, for example, After the femoral stem prosthesis is placed in the medullary cavity, the angle between the long axis of the femoral stem prosthesis and the long axis of the femur is less than or equal to 3°.

Next, referring to FIG. 8 , the above technical solution of the present application will be described as a whole. The preoperative planning method for total hip replacement based on deep learning provided by this application includes:

Step 801: Obtain the three-dimensional block diagram of the total hip joint to be identified.

Step 802: Input the 3D block image of the total hip joint to be identified into the trained 3D segmentation neural network to obtain the femur region in each 2D cross-sectional image of the total hip joint.

Step 803: Based on the three-dimensional reconstruction technology, according to the femoral region in each two-dimensional cross-sectional image of the total hip joint, a three-dimensional image of the femoral region is obtained.

Step 804: Determine the size information of the acetabular cup prosthesis, as well as the type and placement position of the femoral stem prosthesis.

Step 805: Put the determined posterior acetabular cup prosthesis and femoral stem prosthesis into the acetabular fossa and medullary cavity respectively.

Step 806: Correcting the position or type of the inserted acetabular cup prosthesis and the position or type of the femoral stem prosthesis.

Step 807: Output the preoperative planning scheme for total hip replacement.

The following describes the preoperative planning system for total hip replacement based on deep learning provided by this application. The preoperative planning system for total hip replacement based on deep learning described below is the same as the preoperative planning system for total hip replacement based on deep learning described above Planning methods can be cross-referenced.

Fig. 9 is a schematic structural diagram of the preoperative planning system for total hip replacement based on deep learning provided by the present application. As shown in Fig. 9, the present application provides a preoperative planning system for total hip replacement based on deep learning, including A hip joint image collection module 901, a total hip joint recognition module 902 and a total hip joint three-dimensional image construction module 903, wherein the total hip joint image collection module 901 is used to obtain a three-dimensional block diagram of the total hip joint to be identified, and the total hip joint The three-dimensional block diagram of the joint is formed by stacking multiple two-dimensional cross-sectional images of the total hip joint; the total hip joint identification module 902 is used to input the three-dimensional block diagram of the total hip joint to be identified into the trained three-dimensional segmentation In the neural network, the femur region in each two-dimensional cross-sectional image of the total hip joint output by the trained three-dimensional segmentation neural network is obtained, and the trained three-dimensional segmentation neural network is a pre-prepared image of a label marked with the femoral region. Assuming a three-dimensional block diagram, it is obtained by training a convolutional neural network; the three-dimensional image construction module 903 of the total hip joint is used to obtain the described 3D image of the femoral region.

In this application, a two-dimensional cross-sectional image of the hip joint is used for illustration. The two-dimensional cross-sectional image is DICOM data, and each two-dimensional cross-sectional image of the hip joint includes an image of the pelvic region, an image of the left femur, and an image of the right femur. image. Further, the total hip joint image acquisition module 901 converts the DICOM data of the two-dimensional cross-section of the hip joint into a picture in JPG format, and according to the acquisition order of the cross-section (for example, from the femoral head to the end of the femur), multiple converted hip joints The two-dimensional cross-sectional images of the joints are stacked to generate the three-dimensional block diagram of the total hip joint to be identified, that is, the three-dimensional block diagram of the hip joint is obtained.

In this application, the trained three-dimensional segmentation neural network is configured in the total hip joint identification module 902, which can identify the stacked three-dimensional block diagrams of the hip joint. In this application, the convolution kernel of the convolutional neural network used to construct the three-dimensional segmentation neural network is a three-dimensional convolution kernel, so that the three-dimensional segmentation neural network obtained based on the convolutional neural network training can be used for image segmentation. Feature extraction is carried out in three dimensions of the three-dimensional block map of the hip joint, effectively extracting the feature information of the femoral region between each cross-section, and finally segmenting the three-dimensional block map corresponding to the femoral region. Preferably, two continuous three-dimensional convolution kernels in the convolutional neural network are connected through a residual structure; and the loss function of the convolutional neural network is composed of DICE loss and BCE loss.

During the training process of the three-dimensional segmentation neural network, the femur region of the sample hip joint three-dimensional block diagram has been marked, so that the three-dimensional segmentation neural network in the total hip joint recognition module 902 can directly perform femur region analysis on the three-dimensional hip joint image. recognition. It should be noted that, when the present application is training the convolutional neural network, the pixels of the femoral head in the sample femoral area in the three-dimensional block diagram of the sample hip joint are also marked, so that the total hip joint identification module 802 can identify the entire femoral area At the same time, it can further identify the femoral head region, so that the femoral region in the hip joint can be quickly identified, saving time and cost, and improving the recognition accuracy of the hip joint.

Finally, through the total hip joint three-dimensional image construction module 903, the key point information of the identified femoral region is used to perform three-dimensional reconstruction to obtain a three-dimensional image of the femur, so that in the subsequent preoperative planning of total hip joint replacement, the three-dimensional image of the femur can be Determine the size, type and placement of the prosthesis relative to the total hip.

The preoperative planning system for total hip replacement based on deep learning provided by this application stacks multiple two-dimensional cross-sectional images of the total hip joint into a three-dimensional block diagram of the total hip joint, and is based on a convolutional neural network with a three-dimensional convolution kernel structure. Identify the three-dimensional block diagram of the total hip joint, effectively extract the femoral region of each two-dimensional cross-sectional image of the total hip joint, and perform three-dimensional modeling according to the extracted femoral region, so that according to the three-dimensional model of the femoral region, in the Improve the accuracy of total hip joint recognition when performing preoperative planning for total hip arthroplasty.

On the basis of the above embodiments, the system also includes a training module, which is used to input the preset three-dimensional block image set as a training sample into the initial three-dimensional segmentation neural network for training, and obtain a trained three-dimensional segmentation neural network; Wherein, the set of preset three-dimensional block images includes several preset three-dimensional block images obtained by stacking preset two-dimensional cross-sectional images.

On the basis of the above embodiments, the training module includes a sample two-dimensional cross-sectional image acquisition unit, a first labeling unit, a block diagram first construction unit and a first training unit, wherein the sample two-dimensional cross-sectional image acquisition unit It is used to obtain several two-dimensional cross-sectional images of the sample hip joint; the first labeling unit is used to mark the sample femoral area in each sample hip joint two-dimensional cross-sectional image, and to mark the femoral head of the sample femoral area The pixels are labeled with the femoral head area label to obtain several first preset two-dimensional cross-sectional images of the hip joint; The preset two-dimensional cross-sectional images of the hip joint are stacked to obtain a corresponding first preset three-dimensional block image; the first training unit is configured to input a plurality of the first preset three-dimensional block images into the initial three-dimensional Segment the neural network for training to obtain a trained three-dimensional segmentation neural network;

Wherein, the initial three-dimensional segmentation neural network is constructed by the U-Net convolution network, and the convolution kernel of the U-Net convolution network is a three-dimensional convolution kernel.

On the basis of the above-mentioned embodiments, the system further includes a second labeling unit, a second construction unit of the block diagram, and a second training unit, wherein the second labeling unit is used to identify several sheets of the first preset hip joint In the two-dimensional cross-sectional image, the cortical bone pixels of the sample femoral region are labeled with the cortical bone region label to obtain several second preset two-dimensional cross-sectional images of the hip joint; The preset two-dimensional cross-sectional images of the hip joint are stacked in order to obtain the corresponding second preset three-dimensional block diagram; the second training unit is used to train the training The parameters of a good three-dimensional segmentation neural network are optimized to obtain a hip joint recognition model.

On the basis of the above embodiments, the system also includes a cortical bone area identification module, a medullary cavity area determination module, a medullary cavity anatomical axis determination module, a femoral neck-shaft angle calculation module and a femoral stem prosthesis determination module, wherein the cortical bone The area recognition module is used to input the three-dimensional block diagram of the hip joint into the hip joint recognition model if the three-dimensional block diagram of the total hip joint to be identified is a three-dimensional block diagram of the hip joint, and obtain two dimensions of each hip joint. The femoral region and the cortical bone region in the three-dimensional cross-sectional image; the medullary cavity region determination module is used to determine the medullary cavity region according to the femoral region and the cortical bone region; the medullary cavity anatomical axis determination module is used to calculate the medullary cavity The midpoint coordinates of each medullary cavity level in the region, and according to the midpoint coordinates, all the central points are fitted with a straight line to determine the anatomical axis of the medullary cavity; the femoral neck shaft angle calculation module is used to calculate the anatomical axis of the medullary cavity and femoral neck axis to calculate the angle value of the femoral neck shaft angle; the femoral stem prosthesis determination module is used to determine the type of femoral stem prosthesis model and Placement.

On the basis of the foregoing embodiments, the system also includes a femoral head center point pixel coordinate calculation module, a coordinate conversion module, a femoral head rotation center determination module, and an acetabular cup prosthesis determination module, wherein the bone center point pixel coordinate calculation module It is used to obtain the pixel coordinates of the center point of the femoral head in the femoral region in the three-dimensional image based on the centroid formula of the planar image; the coordinate transformation module is used to convert the pixel coordinates into image coordinates; the femoral head rotation center determination module is used to determine The position of the center of rotation of the femoral head; the acetabular cup prosthesis determination module is used to obtain the first size information according to the position of the center of rotation of the femoral head, so as to determine the second size information according to the first size information, wherein the first size information is the size information corresponding to the femoral head, and the second size information is the size information corresponding to the acetabular cup prosthesis model.

On the basis of the above-mentioned embodiments, the preoperative planning system also includes: a position correction module, which is used to adjust the placement position of the placed femoral stem prosthesis model or the placement position of the acetabular cup prosthesis model Correction, so that the placement position of the femoral stem prosthesis model and the placement position of the acetabular cup prosthesis model meet the preset position requirements.

On the basis of the above-mentioned embodiments, the three-dimensional image construction module of the total hip joint is specifically used for:

Determine the intersection point between the center of rotation of the femoral head and the edge of the femoral head area;

The first size information is obtained through the length value between the intersection point and the center position of the femoral head.

On the basis of the above-described embodiments, the three-dimensional image construction module of the total hip joint, after determining the intersection point between the center of rotation of the femoral head and the edge of the femoral head region, obtains the first size information through the length value between the intersection point and the center position of the femoral head when, specifically for;

Determining multiple intersections between the position of the center of rotation of the femoral head and the edge of the femoral head area at different angles;

The first dimension information is obtained by using the average value or the median value of the multiple length values between the multiple intersection points and the central position of the femoral head.

On the basis of the above-mentioned embodiments, the total hip joint recognition module is specifically used for:

The cortical bone area in the femoral area was filtered out to obtain the medullary cavity area.

The system provided in the present application is used to execute the above method embodiments. For the specific process and details, please refer to the above embodiments, which will not be repeated here.

Fig. 10 is a schematic structural diagram of an electronic device provided by the present application. As shown in Fig. 10, the electronic device may include: a processor (Processor) 1001, a communication interface (Communications Interface) 1002, a memory (Memory) 1003 and a communication bus 1004, Wherein, the processor 1001 , the communication interface 1002 , and the memory 1003 communicate with each other through the communication bus 1004 . The processor 1001 can call the logic instructions in the memory 1003 to execute a deep learning-based image planning method for preoperative total hip arthroplasty.

An example, the processor 1001 is configured to obtain a three-dimensional block diagram of the total hip joint to be identified, and the three-dimensional block diagram of the total hip joint to be identified is obtained by stacking multiple two-dimensional cross-sectional images of the total hip joint The three-dimensional block diagram of the total hip joint to be identified is input into the trained three-dimensional segmentation neural network, and each two-dimensional cross-sectional image of the total hip joint output by the trained three-dimensional segmentation neural network is obtained. femoral region, the trained three-dimensional segmentation neural network is obtained by training the convolutional neural network using the preset three-dimensional block diagram marked with the label of the femoral region as a training sample; based on the three-dimensional reconstruction technology, according to each The femoral region in the two-dimensional cross-sectional image of the total hip joint is obtained to obtain a three-dimensional image of the femoral region.

In addition, the above logic instructions in the memory 1003 may be implemented in the form of software function units and when sold or used as an independent product, may be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or the part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including Several instructions are used to make a computer device (which may be a personal computer, a server, or a network device, etc.) execute all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned storage media include: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic disk or optical disc, etc., which can store program codes. .

On the other hand, the present application also provides a computer program product, the computer program product includes a computer program stored on a non-transitory computer-readable storage medium, the computer program includes program instructions, and when the program instructions are executed by a computer During execution, the computer can execute the deep learning-based preoperative planning method for total hip replacement provided by the above methods.

An example is to obtain the three-dimensional block diagram of the total hip joint to be identified, and the three-dimensional block diagram of the total hip joint to be identified is formed by stacking multiple two-dimensional cross-sectional images of the total hip joint; The three-dimensional block diagram of the total hip joint is input into the trained three-dimensional segmentation neural network to obtain the femur region in each two-dimensional cross-sectional image of the total hip joint output by the three-dimensional segmentation neural network trained well. The 3D segmentation neural network is obtained by training the convolutional neural network with the preset 3D block diagram marked with the label of the femoral region as the training sample; based on the 3D reconstruction technology, according to each 2D cross-sectional image of the total hip joint In the femoral region, a three-dimensional image of the femoral region is obtained.

In another aspect, the present application also provides a non-transitory computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it is implemented to perform the deep learning-based total hip joint provided by the above-mentioned embodiments. Preoperative planning methods for replacement surgery.

An example is to obtain the three-dimensional block diagram of the total hip joint to be identified, and the three-dimensional block diagram of the total hip joint to be identified is formed by stacking multiple two-dimensional cross-sectional images of the total hip joint; The three-dimensional block diagram of the total hip joint is input into the trained three-dimensional segmentation neural network to obtain the femur region in each two-dimensional cross-sectional image of the total hip joint output by the three-dimensional segmentation neural network trained well. The 3D segmentation neural network is obtained by training the convolutional neural network with the preset 3D block diagram marked with the label of the femoral region as the training sample; based on the 3D reconstruction technology, according to each 2D cross-sectional image of the total hip joint In the femoral region, a three-dimensional image of the femoral region is obtained.

The device embodiments described above are only illustrative, and the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in One place, or it can be distributed to multiple network elements. Part or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. It can be understood and implemented by those skilled in the art without any creative effort.

Through the above description of the implementations, those skilled in the art can clearly understand that each implementation can be implemented by means of software plus a necessary general-purpose hardware platform, and of course also by hardware. Based on this understanding, the essence of the above technical solution or the part that contributes to the prior art can be embodied in the form of software products, and the computer software products can be stored in computer-readable storage media, such as ROM/RAM, magnetic discs, optical discs, etc., including several instructions to make a computer device (which may be a personal computer, server, or network device, etc.) execute the methods described in various embodiments or some parts of the embodiments.

Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, rather than limiting them; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: it can still Modifications are made to the technical solutions described in the foregoing embodiments, or equivalent replacements are made to some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the various embodiments of the present application. .

Claims (10)

1. A preoperative planning system for total hip replacement based on deep learning, including:

   A total hip joint image acquisition module, configured to acquire a three-dimensional block diagram of the total hip joint to be identified, wherein the three-dimensional block diagram of the total hip joint to be identified is formed by stacking multiple two-dimensional cross-sectional images of the total hip joint;

   The total hip joint identification module is used to input the three-dimensional block diagram of the total hip joint to be identified into the trained three-dimensional segmentation neural network, and obtain the two images of each total hip joint output by the trained three-dimensional segmentation neural network. The femur region in the three-dimensional cross-sectional image, the trained three-dimensional segmentation neural network is obtained by training the convolutional neural network by using the preset three-dimensional block image marked with the label of the femoral region as a training sample;

   The three-dimensional image construction module of the total hip joint is used to obtain a three-dimensional image of the femoral region based on the three-dimensional reconstruction technology based on the femoral region in each two-dimensional cross-sectional image of the total hip joint.
2. According to the deep learning-based preoperative planning system for total hip replacement according to claim 1, the total hip recognition module is also used for:

   Obtain several two-dimensional cross-sectional images of the sample hip joint;

   Annotate the sample femoral region in each sample hip joint two-dimensional cross-sectional image, and mark the femoral head region label on the femoral head pixel of the sample femoral region, and obtain several first preset two-dimensional cross-sectional hip joints image;

   According to the acquisition order of the two-dimensional cross-section of the sample hip joint, stacking several first preset two-dimensional cross-sectional images of the hip joint to obtain a corresponding first preset three-dimensional block image;

   Inputting a plurality of the first preset three-dimensional block images into the initial three-dimensional segmentation neural network for training to obtain a trained three-dimensional segmentation neural network;

   Wherein, the initial three-dimensional segmentation neural network is constructed by the U-Net convolution network, and the convolution kernel of the U-Net convolution network is a three-dimensional convolution kernel.
3. According to the preoperative planning system for total hip arthroplasty based on deep learning according to claim 2, the total hip joint recognition module, when inputting a plurality of the first preset three-dimensional block images into the initial three-dimensional segmentation The neural network is trained, and after the trained three-dimensional segmentation neural network is obtained, it is also used for:

   Marking the cortical bone region labels on the cortical bone pixels of the sample femur region in the first preset two-dimensional cross-sectional images of the hip joint to obtain several second preset two-dimensional cross-sectional images of the hip joint;

   Stacking several second preset hip joint two-dimensional cross-sectional images in order to obtain a corresponding second preset three-dimensional block diagram;

   A hip joint recognition model is obtained by optimizing the parameters of the trained three-dimensional segmentation neural network through a plurality of second preset three-dimensional block diagrams.
4. According to the preoperative planning system for total hip arthroplasty based on deep learning according to claim 3, the total hip joint identification module, after acquiring the three-dimensional block diagram of the joint to be identified, is further used for:

   Inputting the three-dimensional block diagram of the hip joint into the hip joint recognition model to obtain the femoral region and the cortical bone region in each two-dimensional cross-sectional image of the hip joint;

   determining the medullary cavity area according to the femoral area and the cortical bone area;

   calculating the midpoint coordinates of each medullary canal level in the medullary cavity region, and performing straight line fitting on all the midpoints according to the midpoint coordinates to determine the anatomical axis of the medullary cavity;

   Calculate the angle value of the femoral neck-shaft angle according to the anatomical axis of the medullary cavity and the femoral neck axis;

   According to the angle value, the medullary cavity area and the position of the center of rotation of the femoral head, the type and placement position of the femoral stem prosthesis model are determined.
5. According to the preoperative planning system for total hip joint replacement based on deep learning according to any one of claims 1 to 4, the three-dimensional image construction module of the total hip joint, based on the three-dimensional reconstruction technology, according to two images of each total hip joint The femur region in the three-dimensional cross-sectional image, after obtaining the three-dimensional image of the femur region, is also used for:

   Based on the centroid formula of the planar image, obtain the pixel coordinates of the center point of the femoral head in the femoral region in the three-dimensional image;

   converting said pixel coordinates to image coordinates;

   Determine the position of the center of rotation of the femoral head;

   According to the rotation center position of the femoral head, the first size information is obtained to determine the second size information according to the first size information, wherein the first size information is the size information corresponding to the femoral head, and the second size information is Size information corresponding to the acetabular cup prosthesis model.
6. According to the preoperative planning system for total hip arthroplasty based on deep learning according to claim 5, the preoperative planning system also includes: a position correction module, which is used to place the placed femoral stem prosthesis model The position or the placement position of the acetabular cup prosthesis model is corrected, so that the placement position of the femoral stem prosthesis model and the placement position of the acetabular cup prosthesis model meet the preset position requirements.
7. The preoperative planning system for total hip arthroplasty based on deep learning according to claim 5, wherein the three-dimensional image construction module of the total hip joint is specifically used to obtain the first size information according to the position of the center of rotation of the femoral head. :

   Determine the intersection point between the center of rotation of the femoral head and the edge of the femoral head area;

   The first size information is obtained through the length value between the intersection point and the center position of the femoral head.
8. The preoperative planning system for total hip arthroplasty based on deep learning according to claim 7, wherein the three-dimensional image construction module of the total hip joint determines the intersection point between the center of rotation of the femoral head and the edge of the femoral head area, and passes the intersection point and the femoral head area edge. The length value between the center positions of the bones is specifically used when obtaining the first size information;

   Determining multiple intersections between the position of the center of rotation of the femoral head and the edge of the femoral head area at different angles;

   The first dimension information is obtained by using the average value or the median value of the multiple length values between the multiple intersection points and the central position of the femoral head.
9. The preoperative planning system for total hip arthroplasty based on deep learning according to claim 2, wherein two continuous three-dimensional convolution kernels in the convolutional neural network are connected through a residual structure, and the convolution The loss function of the neural network is composed of DICE loss and BCE loss.
10. The preoperative planning system for total hip arthroplasty based on deep learning according to claim 4, wherein the total hip joint identification module, when determining the medullary cavity area according to the femoral area and the cortical bone area, specifically Used for:

    The cortical bone area in the femoral area was filtered out to obtain the medullary cavity area.

Images