Deep learning-based method and system for calculating postoperative offset after total hip arthroplasty

Abstract

The application provides a kind of based on deep learning's calculation method and system of postoperative eccentricity of total hip arthroplasty, it is related to medical technical field, can realize the accurate evaluation of the postoperative condition of the patient of total hip arthroplasty surgery, method includes: obtaining the hip image of the patient after hip arthroplasty surgery;Key point position and target area in the hip image are identified based on the target recognition network of deep learning;According to the key point position and the target area, the eccentricity of the patient is determined.The system executes the method.The application is based on the hip image of the patient after hip arthroplasty surgery, calculates the eccentricity of the patient after hip arthroplasty surgery, to realize the accurate evaluation of the recovery condition of the patient after hip arthroplasty surgery.

Description

TECHNICAL FIELD

[0001] The present application relates to the medical technical field, and in particular to a full hip joint postoperative eccentricity calculation method and system based on deep learning. BACKGROUND

[0002] In the medical field, postoperative evaluation of total hip arthroplasty plays a very important role in the success rate of the operation, so how to provide accurate postoperative evaluation is very important.

[0003] At present, the main preoperative evaluation method is to measure manually through various tools, which is low in efficiency and cannot guarantee accuracy, so it is urgent to provide a more convenient and accurate postoperative evaluation method. SUMMARY

[0004] The full hip joint postoperative eccentricity calculation method and system based on deep learning provided by the present application are used to solve the above problems in the prior art, and the eccentricity of a patient after hip arthroplasty is calculated based on the hip joint image of the patient after the hip arthroplasty, so as to accurately evaluate the recovery of the patient after the hip arthroplasty.

[0005] The full hip joint postoperative eccentricity calculation method based on deep learning provided by the present application comprises the following steps: acquiring a hip joint image of a patient after hip arthroplasty; identifying key point positions and target regions in the hip joint image based on a deep learning target recognition network; and determining the eccentricity of the patient according to the key point positions and the target regions.

[0006] According to the full hip joint postoperative eccentricity calculation method based on deep learning provided by the present application, the target recognition network is trained based on a point recognition neural network and a segmentation neural network, or is trained based on a preset neural network model comprising a stacked hourglass network structure, a segmentation Segment-Head network and a key point Keypoint-Head network.

[0007] According to the application, a method for calculating postoperative eccentricity of total hip arthroplasty based on deep learning is provided. The target recognition network based on deep learning identifies the key point position and target region in the hip image, including: inputting the hip image into the target recognition network to identify the first tear drop point position, the second tear drop point position, the pubic symphysis point position, the femoral prosthesis ball head region, the healthy side femoral head region, the bilateral bone cortex region and the ischium region in the hip image; the first tear drop point position and the second tear drop point position are determined as the first key point position, and the pubic symphysis point position is determined as the second key point position; the key point position is determined according to the first key point position and the second key point position; the femoral prosthesis ball head region, the healthy side femoral head region, the bilateral bone cortex region and the ischium region are determined as the target region.

[0008] According to the application, a method for calculating postoperative eccentricity of total hip arthroplasty based on deep learning is provided. The key point position and the target region are determined to determine the eccentricity of the patient, including: determining the first femoral medullary cavity center line on the same side of the femoral prosthesis ball head region and the second femoral medullary cavity center line on the same side of the healthy side femoral head region according to the bilateral bone cortex region; determining the first shortest distance between the first rotation center point of the femoral prosthesis ball head region and the first femoral medullary cavity center line; determining the second shortest distance between the second rotation center of the healthy side femoral head region and the second femoral medullary cavity center line; determining the femoral eccentricity according to the difference between the first shortest distance and the second shortest distance; wherein the eccentricity includes the femoral eccentricity.

[0009] According to the application, a method for calculating postoperative eccentricity of total hip arthroplasty based on deep learning is provided. The key point position and the target region are determined to determine the eccentricity of the patient, including: determining the first rotation center point of the femoral prosthesis ball head region, the second rotation center point of the healthy side femoral head region, the ischial tuberosity line and the pelvic mid-axis to determine the acetabular cup eccentricity; or, according to the first rotation center point, the second rotation center point, the bilateral tear drop point line and the pelvic mid-axis, the acetabular cup eccentricity is determined; wherein the ischial tuberosity line is determined according to the first lowest point and the second lowest point of the bilateral ischium region; the bilateral tear drop point line is determined according to the first key point position; the pelvic mid-axis is determined according to the second key point position and the ischial tuberosity line; the eccentricity includes the acetabular cup eccentricity.

[0010] According to the application, a method for calculating postoperative eccentricity of total hip arthroplasty based on deep learning is provided, and the postoperative eccentricity of total hip arthroplasty is determined according to a first rotation center point of a femoral head region, a second rotation center point of a healthy side femoral head region, an ischial tuberosity line and a pelvic mid-axis, and includes the following steps: determining a third shortest distance between the first rotation center point and the ischial tuberosity line; determining a fourth shortest distance between the second rotation center point and the ischial tuberosity line; determining a fifth shortest distance between the first rotation center point and the pelvic mid-axis; determining a sixth shortest distance between the second rotation center point and the pelvic mid-axis; and determining the postoperative eccentricity of total hip arthroplasty according to a difference between the third shortest distance and the fourth shortest distance and a difference between the fifth shortest distance and the sixth shortest distance.

[0011] According to the application, a method for calculating postoperative eccentricity of total hip arthroplasty based on deep learning is provided, and the postoperative eccentricity of total hip arthroplasty is determined according to the first rotation center point, the second rotation center point, a line connecting two bilateral tear drop points and the pelvic mid-axis, and includes the following steps: determining a seventh shortest distance between the first rotation center point and the line connecting two bilateral tear drop points; determining an eighth shortest distance between the second rotation center point and the line connecting two bilateral tear drop points; determining a ninth shortest distance between the first rotation center point and the pelvic mid-axis; determining a tenth shortest distance between the second rotation center point and the pelvic mid-axis; and determining the postoperative eccentricity of total hip arthroplasty according to a difference between the seventh shortest distance and the eighth shortest distance and a difference between the ninth shortest distance and the tenth shortest distance.

[0012] The application further provides a system for calculating postoperative eccentricity of total hip arthroplasty based on deep learning, which includes an acquisition module, an identification module and a calculation module; the acquisition module is used to acquire a hip joint image of a patient after hip arthroplasty surgery; the identification module is used to identify key point positions and target regions in the hip joint image based on a deep learning target identification network; and the calculation module is used to determine the eccentricity of the patient according to the key point positions and the target regions.

[0013] The application further provides an electronic device, which includes a memory, a processor and a computer program stored in the memory and executable on the processor, and the processor implements the method for calculating postoperative eccentricity of total hip arthroplasty based on deep learning according to any one of the above when executing the program.

[0014] The application further provides a non-transitory computer readable storage medium, which stores a computer program, and the computer program is executable on a processor to implement the method for calculating postoperative eccentricity of total hip arthroplasty based on deep learning according to any one of the above.

[0015] The application provides a deep learning-based total hip postoperative eccentricity calculation method and system, which is based on a hip image of a patient after hip replacement surgery, and calculates the eccentricity of the patient after the hip replacement surgery, so as to accurately evaluate the recovery of the patient after the hip replacement surgery. BRIEF DESCRIPTION OF DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the application or prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are some embodiments of the application, and other drawings can be obtained by those skilled in the art without creative labor.

[0017] Figure 1 is a flowchart of the deep learning-based total hip postoperative eccentricity calculation method provided by the application;

[0018] Figure 2 is a schematic diagram of the recognized pubic symphysis point in the hip image provided by the application;

[0019] Figure 3 is a schematic diagram of the ischium region in the hip image provided by the application;

[0020] Figure 4 is a structural schematic diagram of the preset neural network model provided by the application;

[0021] Figure 5 is a structural schematic diagram of the target recognition network provided by the application;

[0022] Figure 6 is a schematic diagram of the bilateral femoral medullary cavity center line in the hip image provided by the application;

[0023] Figure 7 is a schematic diagram of the first rotation center of the femoral prosthesis ball head region provided by the application;

[0024] Figure 8 is a schematic diagram of the second rotation center point of the healthy side femoral head region provided by the application;

[0025] Figure 9 is a schematic diagram of the determined femoral eccentricity provided by the application;

[0026] Figure 10 is a position schematic diagram of the bilateral lowest points of the ischium region in the hip image provided by the application;

[0027] Figure 11 is a schematic diagram of the ischial tuberosity line in the hip image provided by the application;

[0028] Figure 12 is a schematic diagram of the midline of the pelvis in the hip joint image provided by the present application;

[0029] Figure 13 is a schematic diagram of the connecting line of the bilateral tear drop points in the hip joint image provided by the present application;

[0030] Figure 14 is one of the schematic diagrams for determining the eccentricity of the acetabular cup provided by the present application;

[0031] Figure 15 is the second schematic diagram for determining the eccentricity of the acetabular cup provided by the present application;

[0032] Figure 16 is a structural schematic diagram of the deep learning-based total hip arthroplasty postoperative eccentricity calculation system provided by the present application;

[0033] Figure 17 is a schematic diagram of the physical structure of the electronic device provided by the present application. DETAILED DESCRIPTION

[0034] In order to make the purpose, technical scheme and advantages of the present application clearer, the technical scheme of the present application will be described clearly and completely below in combination with the drawings in the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0035] Figure 1 is a schematic diagram of the deep learning-based total hip arthroplasty postoperative eccentricity calculation method provided by the present application, as shown in Figure 1 , the method comprises:

[0036] S1, acquiring the hip joint image of the patient after hip replacement surgery;

[0037] S2, identifying the key point position and target area in the hip joint image based on the deep learning-based target recognition network;

[0038] S3, determining the eccentricity of the patient according to the key point position and the target area.

[0039] It should be noted that the execution subject of the above method can be a computer device.

[0040] Optionally, after the hip replacement surgery is completed, the doctor will make postoperative evaluation on the patient based on the hip image of the patient after the hip replacement surgery, and by identifying the key point position and the target region in the hip image of the patient after the hip replacement surgery, the recovery of the patient after the hip replacement surgery is evaluated.

[0041] Firstly, the hip image of the patient after the hip replacement surgery is obtained, specifically, the hip image of the patient can be obtained by X-ray shooting, computed tomography (CT) or magnetic resonance imaging (MRI) on the hip of the patient after the hip replacement surgery.

[0042] Secondly, the key point and the target region of the obtained hip image of the patient after the hip replacement surgery are identified, and the key point position and the target region for postoperative evaluation in the hip image are found, for example, the key point position and the target region can be identified by inputting the hip image into the pre-trained target identification network.

[0043] Finally, the eccentricity of the patient after the surgery is determined according to the identified key point position and target region.

[0044] It should be noted that the obtained eccentricity of the patient after the hip replacement surgery can be used to analyze the accuracy of the installation position of the femoral prosthesis of the patient after the hip replacement surgery, and thus the recovery of the patient after the hip replacement surgery is accurately evaluated.

[0045] The method for calculating the eccentricity of the total hip replacement surgery based on deep learning provided by the application is based on the hip image of the patient after the hip replacement surgery, and the eccentricity of the patient after the hip replacement surgery is calculated to accurately evaluate the recovery of the patient after the hip replacement surgery.

[0046] Further, in one embodiment, the target identification network is trained based on a point identification neural network and a segmentation neural network; or,

[0047] The preset neural network model is trained based on a stack hourglass network structure, a segmentation Segment-Head network and a key point Keypoint-Head network.

[0048] Further, in one embodiment, step S2 can specifically include:

[0049] S21, input the hip joint image into the target recognition network to identify the first tear drop point position, the second tear drop point position, the pubic symphysis point position, the femoral prosthesis ball head region, the healthy side femoral head region, the bilateral bone cortex region and the ischium region on both sides of the ischium region in the hip joint image;

[0050] S22, determine the first tear drop point position and the second tear drop point position as the first key point position, and determine the pubic symphysis point position as the second key point position;

[0051] S23, determine the key point position according to the first key point position and the second key point position;

[0052] S24, determine the femoral prosthesis ball head region, the healthy side femoral head region, the bilateral bone cortex region and the ischium region as the target region.

[0053] Optionally, as shown in the figure, Figures 2-3 the hip joint image of the patient after the hip replacement surgery is input into the pre-trained target recognition network to identify the first tear drop point position, the second tear drop point position, the pubic symphysis point position (such as the G point shown in the figure Figure 2 ) and the target region (including the ischium region, the femoral prosthesis ball head region, the healthy side femoral head region and the bilateral bone cortex region as shown in the figure Figure 3 ) in the ischium region of the hip joint image of the patient after the hip replacement surgery, wherein the target recognition network can be trained by a point recognition neural network and a segmentation neural network, or can be trained by a preset neural network model (including a stacked hourglass network structure (Stacked Hourglass Networks, SHM), a segmentation Segment-Head network and a key point Keypoint-Head network).

[0054] Specifically, the point recognition neural network in the target recognition network can be used to identify the bilateral femoral lesser trochanter lower edge point position, the bilateral tear drop point position and the pubic symphysis point position in the patient's hip joint image labeled in advance, to obtain the first tear drop point position, the second tear drop point position and the pubic symphysis point position on both sides of the ischium region in the patient's postoperative hip joint image; and the segmentation neural network in the target recognition network can be used to convert the hip joint image of the patient after the hip replacement surgery into a 0-255 grayscale image, and classify each pixel point of the image to identify the femoral prosthesis ball head region, the healthy side femoral head region, the bilateral bone cortex region and the ischium region in the hip joint image, for example, each pixel point of the image can be classified according to the ischium region and the background region to determine the ischium region in the hip joint image of the patient after the hip replacement surgery.

[0055] The point recognition neural network can be specifically a target positioning network LocNet, an image segmentation network SegNet, a region convolutional neural network R-CNN, a fast region convolutional neural network Fast R-CNN, a region full convolutional neural network R-FCN, and a target detection network SSD.

[0056] The segmentation neural network can be specifically a full convolutional neural network FCN, a SegNet, a dilated convolutional neural network, an efficient neural network ENet, and a DeepMask instance segmentation network.

[0057] The preset neural network model is trained to obtain the target recognition network, and the specific steps are as follows:

[0058] First, the hip image data set of the patient after the hip replacement surgery is obtained;

[0059] Second, the hip image data set is input into the preset neural network model for training to determine the model output result;

[0060] Finally, the parameters of the preset neural network model are adjusted based on the output result and the loss function until the deep learning model trained is determined;

[0061] The loss function is determined based on the loss function corresponding to the segmentation Segment-Head network and the first weight, and the loss function corresponding to the key point Keypoint-Head network and the second weight.

[0062] It can be understood that before the hip image data set is obtained, the collected hip images of the patient after the hip replacement surgery can be preprocessed. The image format can be a Digital Imaging and Communications in Medicine (DICOM) format file.

[0063] In actual execution, the image format of the hip image of the patient after the hip replacement surgery is first converted to JPG format, and the converted image has the problems of different sizes and diversified contrast.

[0064] For the problem of different sizes, directly scaling the image to the target pixel will cause image distortion and inaccurate subsequent measurement, so the following method can be used for processing: the longer side of the image is scaled to the target pixel ratio, and then the scaled image is zero-filled to avoid the problem of image distortion after conversion. The target pixel can be set to 512x512 pixels.

[0065] For the problem of diversified contrast, the following method can be used for processing:

[0066] I. According to the distribution of pixel values of each image, mean value processing is performed. Then, threshold screening is performed on all images, and contrast enhancement operation is performed on the images with abnormal contrast obtained by screening, so that all images are in the same contrast range.

[0067] II. The image contrast is diversified through gamma transformation, and the data of multiple scenes is increased to adapt to scenes with unknown contrast.

[0068] The above image processing methods can increase image clarity and reduce noise. Of course, in other embodiments, the image processing method can also take other forms, including but not limited to image enhancement using Laplace operator or image enhancement based on object Log transformation, etc. The specific implementation can be determined according to actual needs, and the present application does not make specific limitations thereto.

[0069] For non-DICOM format pictures, the scale of the entire hip joint image is calibrated according to the reference scale on the hip joint image by using deep learning, to ensure the accuracy of subsequent measurement data. For hip joint images with scales, the known size scale can be directly referred to to correct the hip joint image. For hip joint images without scales, the known size of the acetabular cup outer diameter can be referred to to correct the hip joint image.

[0070] Optionally, after the pre-processing operation is completed, a hip joint image data set of a patient after hip replacement surgery can be obtained. The data set includes two parts of key point positions and region segmentation. The key point positions include five key points in each hip joint image, i.e. the first lower edge point position, the second lower edge point position, the first tear drop point position on the bilateral femoral condyles, the second tear drop point position and the pubic symphysis point position. Region segmentation refers to the target segmentation region as the femoral prosthesis ball head region, the healthy side femoral head region, the bilateral bone cortex region and the ischium region. Since the training result needs to be continuously iterated with the true value to reduce the error and improve the prediction accuracy when training the preset neural network model, the hip joint image data set can be divided into a training set, a validation set and a test set according to a target ratio before model training. For example, the target ratio of the training set, the validation set and the test set can be set to 6:2:2.

[0071] Specifically, the deep learning model is built according to different neural network structures, and the training set is input into the preset neural network model for training until each neural network converges, and an initial neural network model is obtained. The initial neural network model is optimized according to the test set, and an optimal neural network model trained is obtained, and the weight parameters of the optimal neural network model are determined. Then, the verification set is input into the optimal neural network model trained for verification, and the output result of the optimal neural network model is verified. In the training process, a multi-weight loss function is used for error calculation, and a back propagation algorithm is used to update the weight parameters of the model constantly until the preset neural network model reaches the expected target, and finally the training is completed.

[0072] Optionally, the loss function in the present application includes two parts, corresponding to the errors of the key point position and the region segmentation result respectively. In order to improve the prediction accuracy of the preset neural network model, the weight changes of the error function corresponding to the key point position and the error function corresponding to the region segmentation are observed during the training process until the errors of the two can be balanced.

[0073] Among them, the loss function corresponds to two different neural network structures and different weights.

[0074] In actual execution, as shown in Figure 4 The network structure of the preset neural network model can include SHM network, Segment-Head network and Keypoint-Head network. The preset neural network model uses Adam optimizer, which combines the advantages of Adagrad (adaptive learning rate gradient descent algorithm) and momentum gradient descent algorithm, which can adapt to sparse gradient (i.e. natural language and computer vision problems) and can also alleviate the problem of gradient shock.

[0075] The loss function of the preset neural network model corresponds to two heads, the loss function of Keypoint-Head is mean absolute error (MAE), that is, the average of the absolute values of the differences between all network prediction points and the corresponding points in the gold standard. The loss function of Segment-Head is Dice coefficient + BCEloss loss function. The total loss function is aMAE+b(Dice+BCEloss), a is the first weight, b is the second weight, which can balance the errors between key points and region segmentation.

[0076] The preset neural network model is evaluated by the following indicators: the evaluation indicator of Keypoints refers to the human key point evaluation indicator ok, and the evaluation indicator of Segment is the Dice coefficient.

[0077] After obtaining the target neural network model, the SHM network and Segment-Head network based on the target neural network model identify the target region in the hip joint images of patients after hip replacement surgery. Taking the ischial region as an example, the details are as follows:

[0078] like Figure 5 As shown, the Hourglass structure is a classic encoder-decoder structure. The encoder consists of convolution and pooling, while the decoder consists of deconvolution and convolution. After the first feature is extracted through the SHM network, the Keypoint-Head and Segment-Head share the feature extraction layer. Based on this, the second feature is further extracted by two convolutions. Finally, the number of channels is changed by a 1×1 convolution, and the output is a logits layer. The Segment-Head extracts the region corresponding to the highest probability value by performing softmax normalization on the logits layer, which is the final segmentation result, i.e., the ischial region.

[0079] The identified femoral prosthesis bulb region, the contralateral femoral head region, the bilateral cortical bone regions, and the ischium region were designated as target regions.

[0080] The location of key points in hip joint images of patients after hip replacement surgery was identified using SHM and Keypoint-Head networks. Specifically:

[0081] like Figure 5 As shown, after the first feature is extracted through the SHM network, Keypoint-Head and Segment-Head share a feature extraction layer. Based on this, the third feature is further extracted through two convolutions. Finally, a 1×1 convolution is used to change the number of channels, and the output is a logits layer. Keypoint-Head generates a heatmap, and the point with the highest probability value in the heatmap is designated as the feature point, i.e., the keypoint (including the first teardrop point, the second teardrop point, and the pubic symphysis point). The first and second teardrop points are determined as the first keypoint locations, and the pubic symphysis point is determined as the second keypoint location. Based on the first and second keypoint locations, the keypoint positions are determined.

[0082] The present invention provides a deep learning-based method for calculating the postoperative eccentricity of total hip arthroplasty, which combines deep learning methods to evaluate the recovery of patients after hip replacement surgery, so as to achieve a rapid and accurate assessment of the postoperative recovery of patients after hip replacement surgery.

[0083] Furthermore, in one embodiment, step S3 may specifically include:

[0084] S31. Based on the bilateral cortical regions, determine the center line of the first medullary cavity on the same side as the femoral prosthesis bulb region and the center line of the second medullary cavity on the same side as the healthy femoral head region;

[0085] S32. Determine the first shortest distance between the first rotation center point of the femoral prosthesis bulb region and the center line of the first femoral medullary cavity.

[0086] S33. Determine the second shortest distance between the second rotation center of the healthy femoral head region and the center line of the second femoral medullary cavity.

[0087] S34. Determine the femoral offset based on the difference between the first shortest distance and the second shortest distance;

[0088] Among them, the eccentricity includes the femoral eccentricity.

[0089] Optionally, after identifying the bilateral cortical bone regions of the hip joint image of a patient after hip replacement surgery, the first medullary canal centerline e1 on the same side as the femoral prosthesis head region and the second medullary canal centerline e2 on the same side as the contralateral femoral head region are calculated, specifically as follows: Figure 6 As shown.

[0090] It should be noted that the femoral medullary cavity centerline is derived by mathematical fitting based on the segmentation of the cortical bone region. First, the image is cut into left and right parts. Then, the segmented cortical bone region is preserved and points are selected at a certain ratio. The point selection method is to group the points with the vertical coordinates of the preserved region, retain the intersection points of the preserved region with the vertical coordinate as the axis, and select the midpoint of the two adjacent points with the largest distance. By traversing all the vertical coordinates of the preserved region, a series of points are obtained. The straight line is obtained by fitting the line using the least squares method, which is the required femoral medullary cavity centerline.

[0091] Based on the identified hip joint images of patients after hip replacement surgery, the first rotation center point F1 of the femoral prosthesis head region and the contralateral femoral head region were calculated (see...). Figure 7 ) and the second rotation center F2 of the contralateral femoral head region (see Figure 8 ).

[0092] The first rotation center F1 of the femoral prosthesis head region is determined by extracting the edge contour of the extracted femoral prosthesis head region using traditional image processing techniques. Three points are taken on the contour, and two straight lines are drawn connecting each pair. The intersection of the perpendicular lines of these two lines is the center of the prosthesis head. The center of the femoral head can then be obtained using the centroid formula for the region of interest. The centroid formula is:

[0093]

[0094] Wherein, the coordinate of each pixel in the image in the x direction is x i , and the corresponding pixel value is P i , the coordinate of the centroid in the x direction is x0, and the coordinate of each pixel in the image in the y direction is y j , and the corresponding pixel value is P i , the coordinate of the centroid in the x direction is y0, and n represents the number of image pixels.

[0095] As shown in Figure 9 , it is assumed that the first rotation center point F1 of the femoral prosthesis ball head region, the second rotation center F2 of the healthy femoral head region, the first femoral medullary cavity center line e1, and the second femoral medullary cavity center line e2.

[0096] A perpendicular line is drawn from the first rotation center point F1 to the first femoral medullary cavity center line e1 to obtain a first line segment F1d1, and a perpendicular line is drawn from the second rotation center point F2 to the first femoral medullary cavity center line e2 to obtain a second line segment F2d2. The first line segment F1d1 is the first shortest distance between the first rotation center point F1 and the first femoral medullary cavity center line e1, and the second line segment F2d2 is the second shortest distance between the second rotation center point F2 and the second femoral medullary cavity center line e2.

[0097] According to the distance of the first line segment F1d1 and the distance of the second line segment F2d2, the difference between the first line segment F1d1 and the second line segment F2d2 is calculated, which is the femoral offset of the patient. The femoral offset can be used to determine the patient's offset, and then the patient's lower limb leg length recovery after joint replacement surgery is determined.

[0098] According to the femoral offset obtained above, the postoperative recovery of the patient undergoing hip replacement surgery is evaluated. If the femoral offset is within a predetermined range (for example, 31mm-45mm), it is determined that the patient undergoing hip replacement surgery has good postoperative recovery.

[0099] This eccentric structure of the femur affects the strength and movement efficiency of the hip abductor muscle. A suitable femoral offset can balance the hip abductor muscle strength, achieve maximum abduction strength and minimum joint interface stress, i.e. pelvic balance can be achieved with minimum abduction muscle strength. With the increase of the offset, the corresponding abduction muscle strength arm increases, the abduction muscle strength decreases, the joint contact stress decreases, the prosthesis wear is reduced, the stress on the neck of the prosthesis is reduced, and the stress on the corresponding part of the femur is reduced.

[0100] The application provides a deep learning-based calculation method for postoperative eccentricity of total hip arthroplasty, which uses a deep learning method to identify and calculate the corresponding key points and target regions in the hip joint image of the hip joint of a patient after total hip arthroplasty, thereby laying a foundation for subsequent rapid evaluation of the postoperative recovery of the patient who has undergone hip arthroplasty based on the femoral eccentricity.

[0101] Further, in one embodiment, step S3 can further specifically include:

[0102] S35, determining the acetabular cup eccentricity according to the first rotation center point of the femoral head region of the femoral prosthesis, the second rotation center point of the contralateral femoral head region, the ischial tuberosity line and the pelvic mid-axis line; or

[0103] S36, determining the acetabular cup eccentricity according to the first rotation center point, the second rotation center point, the line connecting the bilateral tear drop points and the pelvic mid-axis line.

[0104] The ischial tuberosity line is determined according to the bilateral first lowest point and the second lowest point of the ischial region.

[0105] The line connecting the bilateral tear drop points is determined according to the first key point position.

[0106] The pelvic mid-axis line is determined according to the second key point position and the ischial tuberosity line.

[0107] The eccentricity includes the acetabular cup eccentricity.

[0108] Optionally, after identifying the ischial region of the hip joint image of the patient after hip arthroplasty, the bilateral first lowest point and the second lowest point of the ischial region are determined to obtain the ischial tuberosity line, specifically:

[0109] The bilateral lowest points are extracted from the segmented ischial region by using image processing technology, that is, the lowest points of the bilateral ischial regions are taken as the first lowest points, and a horizontal straight line is drawn along the first lowest point, as shown in Figure 10 .

[0110] Then, the horizontal straight line obtained above is rotated around the first lowest point (counterclockwise rotation when the lowest point is on the left side, and clockwise rotation when the lowest point is on the right side) until a second intersection point with the ischial region is obtained, that is, the bilateral second lowest point of the ischial region, as shown in Figure 11 . The bilateral lowest points and the intersection point are connected to obtain the ischial tuberosity line CD.

[0111] Alternatively, after obtaining the ischial region, the ischial edge point set of the ischial region is determined. Each row of pixel points of the ischial region is automatically scanned. The scanning mode is as follows:

[0112] Step 1, scan from the bottom of the ischium region upwards by a horizontal scanning line, and determine whether the scanning line passes through a pixel point on the edge of the ischium every time the pixel point is raised. In the case that the scanning line passes through a first pixel point on the edge of the ischium for the first time, the scanning line stops moving upwards. Or in the case that the point on the scanning line exists in the point set on the edge of the ischium, the scanning line stops moving upwards and determines the first pixel point, assuming that the first pixel point is the first lowest point on both sides of the ischium region.

[0113] Step 2, taking the first pixel point as the center of rotation, determine whether the scanning line passes through a pixel point on the edge of the ischium every time the scanning line is rotated by one degree. In the case that the scanning line passes through a second pixel point on the edge of the ischium for the first time, the scanning line stops rotating. Or in the case that the point on the scanning line exists in the point set on the edge of the ischium, the scanning line stops moving upwards and determines the second pixel point, then the second pixel point is the second lowest point on both sides of the ischium region.

[0114] Step 3, the line connecting the first pixel point and the second pixel point is determined as the greater trochanter line CD.

[0115] According to the first rotation center point F1 of the femoral head region, the second rotation center point F2 of the healthy side femoral head region, the greater trochanter line CD and the pelvic mid-axis EF, the offset of the acetabular cup is determined; or according to the first rotation center point F1, the second rotation center point F2, the line connecting the first and second tear drop points and the pelvic mid-axis EF, the offset of the acetabular cup is determined, which can be used to determine the offset of the patient. The line connecting the first and second tear drop points is obtained by connecting the first and second tear drop points.

[0116] The method for calculating the offset of the full hip joint after the operation based on deep learning provided by the application can identify the corresponding key points in the hip joint image of the patient after the hip replacement surgery and evaluate the offset of the acetabular cup, which lays a foundation for determining the femoral prosthesis index based on the offset of the acetabular cup and realizing the rapid and accurate evaluation of the postoperative recovery of the patient.

[0117] Further, in one embodiment, step S35 can specifically include:

[0118] S351, determining a third shortest distance between the first rotation center point and the greater trochanter line;

[0119] S352, determining a fourth shortest distance between the second rotation center point and the greater trochanter line;

[0120] S353, determining a fifth shortest distance between the first rotation center point and the pelvic mid-axis;

[0121] S354, determining a sixth shortest distance between the second rotation center point and the pelvic mid-axis;

[0122] S355. Determine the acetabular cup offset based on the difference between the third and fourth shortest distances and the difference between the fifth and sixth shortest distances.

[0123] Furthermore, in one embodiment, step S36 may specifically include:

[0124] S361. Determine the seventh shortest distance between the first rotation center point and the line connecting the two teardrop points;

[0125] S362. Determine the eighth shortest distance between the second rotation center point and the line connecting the two teardrop points;

[0126] S363. Determine the ninth shortest distance between the first rotation center point and the pelvic midline.

[0127] S364. Determine the tenth shortest distance between the second rotation center point and the pelvic midline.

[0128] S365. Determine the acetabular cup offset based on the difference between the seventh and eighth shortest distances and the difference between the ninth and tenth shortest distances.

[0129] Optionally, such as Figure 14 As shown, the acetabular cup offset is determined based on the first rotation center point F1, the second rotation center point F2, the ischial tuberosity line CD, and the pelvic midline EF. Specifically:

[0130] Draw perpendicular lines from the first rotation center point F1 and the second rotation center point F2 to the ischial tuberosity line CD, respectively, to obtain the third line segment F1L1 and the fourth line segment F2L2. The distance between the third line segment F1L1 is the third shortest distance between the first rotation center point F1 and the ischial tuberosity line CD, and the distance between the fourth line segment F2L2 is the fourth shortest distance between the second rotation center point F2 and the ischial tuberosity line CD.

[0131] Draw perpendicular lines from the first rotation center point F1 and the second rotation center point F2 to the pelvic midline EF, respectively, to obtain the fifth line segment F1N1 and the sixth line segment F2N2. The distance between the fifth line segment F1N1 is the fifth shortest distance between the first rotation center point F1 and the pelvic midline EF, and the distance between the sixth line segment F2N2 is the sixth shortest distance between the second rotation center point F2 and the pelvic midline EF.

[0132] The difference between the third shortest distance and the fourth shortest distance and the difference between the fifth shortest distance and the sixth shortest distance are calculated, and the acetabular cup eccentricity is determined according to the absolute value of the difference between the third shortest distance and the fourth shortest distance and the absolute value of the difference between the fifth shortest distance and the sixth shortest distance, and the accuracy of the femoral prosthesis installation position is determined according to the calculated acetabular cup eccentricity, for example, if the difference between the absolute value of the difference between the third shortest distance and the fourth shortest distance and the absolute value of the difference between the fifth shortest distance and the sixth shortest distance is within a preset threshold range, it is determined that the accuracy of the femoral prosthesis installation position is high.

[0133] It should be noted that the pelvic mid-axis EF is determined by drawing a perpendicular line to the ischial tuberosity line CD along the pubic symphysis point G (see Figure 12 ), and the bilateral tear drop point connecting line ab is obtained by connecting the first tear drop point D1 and the second tear drop point in the first key point position (see Figure 13 ).

[0134] As shown in Figure 15 , the acetabular cup eccentricity is determined according to the first rotation center point F1, the second rotation center point F2, the bilateral tear drop point connecting line ab and the pelvic mid-axis EF, specifically:

[0135] A perpendicular line is drawn from the first rotation center point F1 and the second rotation center point F2 to the bilateral tear drop point connecting line ab to obtain the seventh line segment F1P1 and the eighth line segment F2P2, the distance between the seventh line segment F1P1 is the seventh shortest distance between the first rotation center point F1 and the bilateral tear drop point connecting line ab, and the distance between the eighth line segment F2P2 is the eighth shortest distance between the second rotation center point F2 and the bilateral tear drop point connecting line ab.

[0136] A perpendicular line is drawn from the first rotation center point F1 and the second rotation center point F2 to the pelvic mid-axis EF to obtain the ninth line segment F1Q1 and the tenth line segment F2Q2, the distance between the ninth line segment F1Q1 is the ninth shortest distance between the first rotation center point F1 and the pelvic mid-axis EF, and the distance between the tenth line segment F2Q2 is the tenth shortest distance between the second rotation center point F2 and the pelvic mid-axis EF.

[0137] The difference between the seventh shortest distance and the eighth shortest distance and the difference between the ninth shortest distance and the tenth shortest distance are calculated, and the acetabular cup eccentricity is calculated according to the absolute value of the difference between the seventh shortest distance and the eighth shortest distance and the absolute value of the difference between the ninth shortest distance and the tenth shortest distance, for example, if the difference between the absolute value of the difference between the seventh shortest distance and the eighth shortest distance and the absolute value of the difference between the ninth shortest distance and the tenth shortest distance is within a preset threshold range, it is determined that the accuracy of the femoral prosthesis installation position is higher.

[0138] The full hip joint postoperative eccentricity calculation method based on deep learning provided by the application identifies corresponding key points in the hip joint image of the patient after the hip replacement surgery and calculates the acetabular cup eccentricity, which lays a foundation for subsequent determination of the femoral prosthesis index based on the acetabular cup eccentricity and rapid and accurate evaluation of the postoperative recovery of the patient.

[0139] The full hip joint postoperative eccentricity calculation system based on deep learning provided by the application is described below, and the full hip joint postoperative eccentricity calculation system based on deep learning described below can be correspondingly referred to the full hip joint postoperative eccentricity calculation method based on deep learning described above.

[0140] Figure 16 The full hip joint postoperative eccentricity calculation system based on deep learning provided by the application is described below, and the full hip joint postoperative eccentricity calculation system based on deep learning described below can be correspondingly referred to the full hip joint postoperative eccentricity calculation method based on deep learning described above. Figure 16 As shown in FIG. 1, the full hip joint postoperative eccentricity calculation system based on deep learning provided by the application comprises:

[0141] The acquisition module 1610, the identification module 1611 and the calculation module 1612;

[0142] The acquisition module 1610 is configured to acquire the hip joint image of the patient after the hip replacement surgery.

[0143] The identification module 1611 is configured to identify the key point position and the target region in the hip joint image based on a target identification network of deep learning.

[0144] The calculation module 1612 is configured to determine the eccentricity of the patient according to the key point position and the target region.

[0145] The full hip joint postoperative eccentricity calculation system based on deep learning provided by the application is based on the hip joint image of the patient after the hip replacement surgery, and calculates the eccentricity of the patient after the hip replacement surgery, so as to realize accurate evaluation of the recovery of the patient after the hip replacement surgery.

[0146] Further, in one embodiment, the identification module 1611 can be specifically used for:

[0147] input the hip joint image into a target recognition network to identify a first tear drop point position, a second tear drop point position, a pubic symphysis point position, a femoral prosthesis ball head region, a healthy side femoral head region, a bilateral bone cortex region and an ischium region in the hip joint image;

[0148] determine the first tear drop point position and the second tear drop point position as first key point positions, and determine the pubic symphysis point position as a second key point position;

[0149] determine a key point position according to the first key point position and the second key point position;

[0150] determine the femoral prosthesis ball head region, the healthy side femoral head region, the bilateral bone cortex region and the ischium region as target regions;

[0151] The target recognition network is trained based on a point recognition neural network and a segmentation neural network.

[0152] The target recognition network is trained based on a preset neural network model including a stacked hourglass network structure, a segmentation Segment-Head network and a key point Keypoint-Head network.

[0153] The application provides a deep learning-based total hip arthroplasty postoperative eccentricity calculation system, which combines a deep learning method to evaluate the recovery of patients after hip arthroplasty, so as to realize rapid and accurate evaluation of the postoperative recovery of patients after hip arthroplasty.

[0154] Further, in one embodiment, the calculation module 1612 can also be specifically used for: determining a first femoral medullary cavity center line on the same side of the femoral prosthesis ball head region and a second femoral medullary cavity center line on the same side of the healthy side femoral head region according to the bilateral bone cortex region;

[0155] determine a first shortest distance between a first rotation center point of the femoral prosthesis ball head region and the first femoral medullary cavity center line;

[0156] determine a second shortest distance between a second rotation center point of the healthy side femoral head region and the second femoral medullary cavity center line;

[0157] determine a femoral eccentricity according to a difference between the first shortest distance and the second shortest distance;

[0158] The eccentricity includes the femoral eccentricity.

[0159] The application provides a deep learning-based total hip postoperative offset calculation system, which utilizes a deep learning method to identify corresponding key points and target regions in a hip image of a patient after total hip replacement surgery and calculate femoral offset, thereby laying a foundation for subsequent rapid evaluation of postoperative recovery of the patient after the total hip replacement surgery based on the femoral offset.

[0160] Further, in one embodiment, the calculation module 1612 can also be specifically used for:

[0161] determining the acetabular cup offset according to the first rotation center point of the femoral head region, the second rotation center point of the contralateral femoral head region, the ischial tuberosity line and the pelvic mid-axis; or

[0162] determining the acetabular cup offset according to the first rotation center point, the second rotation center point, the bilateral tear drop point line and the pelvic mid-axis;

[0163] wherein the ischial tuberosity line is determined according to the bilateral first lowest point and the second lowest point of the ischial region;

[0164] the bilateral tear drop point line is determined according to the first key point position;

[0165] the pelvic mid-axis is determined according to the second key point position and the ischial tuberosity line;

[0166] the offset includes the acetabular cup offset.

[0167] The deep learning-based total hip postoperative offset calculation system provided by the application identifies corresponding key points in a hip image of a patient after total hip replacement surgery and evaluates the acetabular cup offset, thereby laying a foundation for subsequent rapid and accurate evaluation of postoperative recovery of the patient based on the acetabular cup offset to determine the femoral prosthesis index.

[0168] Further, in one embodiment, the calculation module 1612 can also be specifically used for:

[0169] determining a third shortest distance between the first rotation center point and the ischial tuberosity line;

[0170] determining a fourth shortest distance between the second rotation center point and the ischial tuberosity line;

[0171] determining a fifth shortest distance between the first rotation center point and the pelvic mid-axis;

[0172] determining a sixth shortest distance between the second rotation center point and the pelvic mid-axis;

[0173] The acetabular cup eccentricity is determined according to a difference between the third shortest distance and the fourth shortest distance and a difference between the fifth shortest distance and the sixth shortest distance.

[0174] Further, in one embodiment, the calculating module 1612 can be further specifically used for:

[0175] determining a seventh shortest distance between the first rotation center point and a line connecting the bilateral tear drop points;

[0176] determining an eighth shortest distance between the second rotation center point and the line connecting the bilateral tear drop points;

[0177] determining a ninth shortest distance between the first rotation center point and the mid-axis of the pelvis;

[0178] determining a tenth shortest distance between the second rotation center point and the mid-axis of the pelvis;

[0179] The acetabular cup eccentricity is determined according to a difference between the seventh shortest distance and the eighth shortest distance and a difference between the ninth shortest distance and the tenth shortest distance.

[0180] The calculation system for the postoperative eccentricity of the total hip arthroplasty based on deep learning provided by the application lays a foundation for determining the femoral prosthesis index based on the acetabular cup eccentricity and realizing the rapid and accurate evaluation of the postoperative recovery of the patient.

[0181] Figure 17 is a schematic diagram of an entity structure of an electronic device provided by the application, as shown in Figure 17 The electronic device can include a processor 1710, a communication interface 1711, a memory 1712 and a bus 1713, wherein the processor 1710, the communication interface 1711 and the memory 1712 complete the communication with each other through the bus 1713. The processor 1710 can call the logical instructions in the memory 1712 to execute the following method: obtaining the hip joint image of the patient after the hip arthroplasty; identifying the key point position and the target area in the hip joint image based on the target recognition network of deep learning; and determining the eccentricity of the patient according to the key point position and the target area.

[0182] In addition, the logic instructions in the above-mentioned memory can be realized in the form of a software function unit and sold or used as an independent product, and can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or the parts that contribute to the prior art or parts of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium, and includes a plurality of instructions for causing a computer power supply screen (which can be a personal computer, a server, or a network power supply screen) to execute all or part of the steps of the method described in each embodiment of the present application. The foregoing storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various media that can store program codes.

[0183] Further, the present application discloses a computer program product, the computer program product comprising a computer program stored on a non-transitory computer-readable storage medium, the computer program comprising program instructions that, when executed by a computer, cause the computer to perform the deep learning-based calculation method of postoperative eccentricity of total hip arthroplasty provided by the above-mentioned method embodiments, for example, comprising: obtaining a hip joint image of a patient after a hip replacement surgery; identifying the key point position and the target region in the hip joint image based on a deep learning-based target recognition network; and determining the eccentricity of the patient according to the key point position and the target region.

[0184] On the other hand, the present application also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a deep learning-based calculation method of postoperative eccentricity of total hip arthroplasty provided by the above-mentioned embodiments, for example, comprising: obtaining a hip joint image of a patient after a hip replacement surgery; identifying the key point position and the target region in the hip joint image based on a deep learning-based target recognition network; and determining the eccentricity of the patient according to the key point position and the target region.

[0185] The system embodiments described above are only schematic, wherein the units illustrated as separate components can or can not be physically separated, and the components illustrated as units can or can not be physical units, i.e., can be located in one place, or can be distributed on a plurality of network units. Part or all of the modules can be selected according to actual needs to achieve the purpose of the present embodiment. Those skilled in the art can understand and implement without creative labor.

[0186] Those skilled in the art can clearly understand the technical solutions of the various embodiments from the above description of the embodiments, and the various embodiments can be implemented by means of software with the necessary general hardware platforms, and of course, can also be implemented by hardware. Based on such understanding, the above technical solutions, essentially or in other words, the part of the prior art that makes a contribution, can be embodied in the form of a software product, which can be stored in a computer readable storage medium, such as a ROM / RAM, a magnetic disk, an optical disk, and the like, and includes a number of instructions for causing a computer power supply screen (which can be a personal computer, a server, or a network power supply screen, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0187] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for some 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

1. A method for calculating postoperative offset of total hip arthroplasty based on deep learning, characterized in that, The method comprises: obtaining a hip joint image of a patient after hip joint replacement surgery; identifying a key point position and a target region in the hip joint image based on a deep learning-based target recognition network; determining the offset of the patient according to the key point position and the target region; the target recognition network is trained based on a point recognition neural network and a segmentation neural network, or based on a preset neural network model comprising a stacked hourglass network structure, a segmentation Segment-Head network and a key point Keypoint-Head network; the deep learning-based target recognition network identifies the key point position and the target region in the hip joint image, comprising: inputting the hip joint image into the target recognition network to identify the first tear drop point position, the second tear drop point position, the pubic symphysis point position, the femoral prosthesis ball head region, the healthy side femoral head region, the bilateral bone cortex region and the ischium region in the hip joint image; determining the first tear drop point position and the second tear drop point position as the first key point position, and determining the pubic symphysis point position as the second key point position; determining the key point position according to the first key point position and the second key point position; determining the femoral prosthesis ball head region, the healthy side femoral head region, the bilateral bone cortex region and the ischium region as the target region; the key point position is determined based on the following method: extracting the first feature of the hip joint image using the stacked hourglass network structure; based on the first feature, after the feature extraction layer shared by Keypoint-Head and Segment-Head, two convolution operations are performed respectively to extract the third feature, and the logits layer is output by 1x1 convolution operation; based on the Keypoint-Head, the logits layer is used to generate a heat map, and the maximum probability value point in the heat map is extracted as a key point to obtain the key point position. 2.The deep learning-based method of calculating postoperative offset after total hip arthroplasty according to claim 1, wherein, determining the offset of the patient according to the key point position and the target region, comprising: determining the first femoral medullary cavity center line on the same side of the femoral prosthesis ball head region and the second femoral medullary cavity center line on the same side of the healthy side femoral head region according to the bilateral bone cortex region; determining the first shortest distance between the first rotation center point of the femoral prosthesis ball head region and the first femoral medullary cavity center line; determining the second shortest distance between the second rotation center point of the healthy side femoral head region and the second femoral medullary cavity center line; determining the femoral offset according to the difference between the first shortest distance and the second shortest distance; wherein the offset comprises the femoral offset. 3.The method of claim 1, wherein, determining the offset of the patient according to the key point position and the target region, further comprising: determining the acetabular cup offset according to the first rotation center point of the femoral prosthesis ball head region, the second rotation center point of the healthy side femoral head region, the ischial tuberosity line and the pelvic mid-axis; or determine the cup offset according to the first rotation center point, the second rotation center point, a line connecting the bilateral tear drop points, and the mid-axis of the pelvis; wherein the ischial tuberosity line is determined according to the first lowest point and the second lowest point of the bilateral ischial regions; the line connecting the bilateral tear drop points is determined according to the first key point position; the mid-axis of the pelvis is determined according to the second key point position and the ischial tuberosity line; the offset includes the cup offset.

4. The deep learning-based method of calculating post-operative offset for total hip arthroplasty of claim 3, wherein, determine the cup offset according to the first rotation center point, the second rotation center point, a line connecting the bilateral tear drop points, and the mid-axis of the pelvis; determine a third shortest distance between the first rotation center point and the ischial tuberosity line; determine a fourth shortest distance between the second rotation center point and the ischial tuberosity line; determine a fifth shortest distance between the first rotation center point and the mid-axis of the pelvis; determine a sixth shortest distance between the second rotation center point and the mid-axis of the pelvis; determine the cup offset according to the difference between the third shortest distance and the fourth shortest distance, and the difference between the fifth shortest distance and the sixth shortest distance.

5. The deep learning-based method of calculating post-operative offset for total hip arthroplasty of claim 3, wherein, determine the cup offset according to the first rotation center point, the second rotation center point, a line connecting the bilateral tear drop points, and the mid-axis of the pelvis; determine a seventh shortest distance between the first rotation center point and the line connecting the bilateral tear drop points; determine an eighth shortest distance between the second rotation center point and the line connecting the bilateral tear drop points; determine a ninth shortest distance between the first rotation center point and the mid-axis of the pelvis; determine a tenth shortest distance between the second rotation center point and the mid-axis of the pelvis; determine the cup offset according to the difference between the seventh shortest distance and the eighth shortest distance, and the difference between the ninth shortest distance and the tenth shortest distance.

6. A deep learning-based system for calculating postoperative offset after total hip arthroplasty, comprising: comprise: an acquisition module, an identification module, and a calculation module; the acquisition module is configured to acquire a hip joint image of a patient after a hip arthroplasty surgery; the identification module is configured to identify a key point position and a target region in the hip joint image based on a target identification network of deep learning; the calculation module is configured to determine an offset of the patient according to the key point position and the target region; the target identification network is trained based on a point identification neural network and a segmentation neural network, or is trained based on a preset neural network model comprising a stacked hourglass network structure, a segmentation Segment-Head network, and a key point Keypoint-Head network; the identification module is configured to input the hip joint image into the target identification network to identify a first tear drop point position, a second tear drop point position, a pubic symphysis point position, a femoral prosthesis head region, a contralateral femoral head region, bilateral bone cortex regions, and ischial regions of the bilateral ischial regions in the hip joint image. The first tear drop point position and the second tear drop point position are determined as first key point positions, and the pubic symphysis point position is determined as a second key point position; The key point positions are determined according to the first key point positions and the second key point positions; The femoral prosthesis ball head region, the contralateral femoral head region, the bilateral bone cortex region and the ischium region are determined as the target regions; The key point positions are determined in the following manner: The first feature of the hip joint image is extracted by using the stacked hourglass network structure; Based on the first feature, after the feature extraction layer shared by the Keypoint-Head and the Segment-Head, the third feature is extracted by two convolution operations respectively, and the logits layer is output by the 1x1 convolution operation; Based on the Keypoint-Head, the logits layer is used to generate a heat map, the maximum probability value point in the heat map is extracted as a key point, and the key point positions are obtained.

7. An electronic device comprising a processor and a memory having a computer program stored therein, characterized in that The processor executes the computer program to realize the calculation method of the postoperative eccentricity of the total hip arthroplasty based on deep learning according to any one of claims 1 to 5.

8. A non-transitory computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to realize the calculation method of the postoperative eccentricity of the total hip arthroplasty based on deep learning according to any one of claims 1 to 5.

Images