An artificial intelligence joint planing device based on image processing

Abstract

The present invention discloses an artificial intelligence joint planing device based on image processing. According to the construction of key-value pairs and the matching of user physiological information, the features of the constructed planing path on each path are saved. The path conditions during the process and the changes in the user's physiological information are found through a time convolutional network. According to the Bayesian network, the connection mode of the network is designed under the correlation relationship between the path, the three-dimensional image of the path including the internal structure of the joint and the user's physiological information, so that the detection features can be obtained more accurately, and the path is reconstructed according to the detection features and time features according to the position to be planed. The path of the entire process from the approach to the planing of the joint part is obtained. The technical effect of improving the planing precision and the accuracy of the planing during the joint surgery is achieved.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to an artificial intelligence joint planing device based on image processing. Background Art

[0002] At present, the planing equipment used in joints can remove debris outside the joint through an aspirator while cutting tissue. Planing heads and grinding drill bits have different shapes, lengths and diameters, and some planing heads can be bent to meet different surgical space requirements. Arthroscopy has multiple access methods and multiple movement paths for different joint structures. Different locations that need to be planed have different characteristics. If only the same method is used or the path is not predicted, it will cause damage to the joint. Therefore, it is necessary to analyze the characteristics of the location through the images collected by the arthroscopy and the physiological characteristic data, so as to plan it. Summary of the invention

[0003] The purpose of the present invention is to provide an artificial intelligence joint planing device based on image processing to solve the above-mentioned problems existing in the prior art.

[0004] In a first aspect, an embodiment of the present invention provides an artificial intelligence joint planing device based on image processing, including a processor;

[0005] The processor is used to execute the following method:

[0006] Acquire arthroscopic images and user physiological information at multiple time points; the user physiological information represents the physiological information corresponding to the user when the arthroscopic images at multiple time points are acquired;

[0007] Based on the arthroscopic images at the multiple time points, determine the internal structure of the joint, match the positions in the images, and obtain a three-dimensional image of the path;

[0008] Based on the three-dimensional image of the path, the path is projected into a graph in the data structure to obtain a path structure graph; the path structure graph is a graph represented by a point set and an edge set;

[0009] Matching the user's physiological information to the path structure graph to obtain a matching graph;

[0010] Through the Bayesian network and the time convolution network, based on the matching graph and the three-dimensional image of the path, the path of the inserted planing device and the joint part to be planed are analyzed to obtain the planing path; the planing path represents the path completed from the inserted joint to the key part to be planed.

[0011] Optionally, the step of analyzing the path of the planing device and the joint part to be planed based on the matching graph and the three-dimensional image of the path by using a Bayesian network and a temporal convolutional network to obtain the planing path includes:

[0012] Marking the area corresponding to the edge set corresponding to the matching graph in the path three-dimensional image to obtain a marked three-dimensional image; obtaining multiple marked three-dimensional images corresponding to multiple path sets;

[0013] Arrange the key-value pairs corresponding to the edge set in the path set according to the connectivity order of the point set in the path set to obtain a path vector;

[0014] Inputting the path vector into a path time convolution network, extracting features of line segment direction, line segment length and user physiological information at different time points, and obtaining time features;

[0015] Based on the multiple labeled three-dimensional images and path vectors, a Bayesian network is constructed to extract features, perform path selection, and obtain Bayesian features;

[0016] The Bayesian features and time features are input into the deconvolution network to determine the damaged joint position, reconstruct the image, and obtain the planing path.

[0017] Optionally, projecting the path into a graph in a data structure based on the path three-dimensional image to obtain a path structure graph includes:

[0018] Based on the path three-dimensional image, a plurality of partial path skeleton point sets are obtained; the partial path skeleton point sets are sets obtained by segmenting the plurality of path skeleton points obtained from the path three-dimensional image;

[0019] Based on the partial path skeleton point set, a plurality of path straight lines are obtained;

[0020] The intersection of the path straight lines corresponding to two adjacent partial path skeleton point sets is taken as the mutation point;

[0021] Mark the two mutation points on the path line according to the connection position to obtain the path segment start point and the path segment end point;

[0022] The line segment between the starting point and the end point of the path segment on the path straight line is constructed as the path segment; and the line segment direction and the line segment length are obtained according to the starting point and the end point of the path segment;

[0023] A path structure diagram is obtained based on the path segment starting point, the path segment end point, the path segment, the segment direction and the segment length.

[0024] Optionally, constructing a Bayesian network based on the multiple labeled three-dimensional images and the path vectors, extracting features, performing path selection, and obtaining Bayesian features includes:

[0025] Inputting the labeled three-dimensional image into a first convolutional network to obtain a first path feature; obtaining a plurality of first path features corresponding to a plurality of labeled three-dimensional images;

[0026] Obtaining the accuracy of a second convolutional network that uses the first path feature and the line segment direction and line segment length in the path vector as input values ​​as a first accuracy;

[0027] obtaining the user physiological information in the path vector and the first path feature as an accuracy of a third convolutional network as input values ​​as a second accuracy;

[0028] Normalizing the first accuracy and the second accuracy to obtain a first normalized value and a second normalized value;

[0029] According to the first normalized value and the second normalized value as probabilities, a Bayesian network is constructed by using the first path feature, the line segment direction and line segment length in the path vector, and the user physiological information in the path vector;

[0030] According to the Bayesian network, the second convolutional network and the third convolutional network are connected to obtain a detection network;

[0031] Inputting the labeled three-dimensional image and the path vector into a trained detection network to obtain detection features; multiple labeled three-dimensional images correspond to multiple detection features;

[0032] Input the detection feature into a fully connected network, detect the path, and obtain a detection value; multiple detection values ​​are obtained corresponding to multiple detection features;

[0033] The detection feature corresponding to the detection value greater than other detection values ​​among the multiple detection values ​​is used as the Bayesian feature.

[0034] Optionally, judging the internal structure of the joint based on the arthroscopic images at the multiple time points, matching the positions in the images, and obtaining the three-dimensional path image includes:

[0035] Gray-scaling the arthroscopic image to obtain an arthroscopic grayscale image;

[0036] The movement direction, movement distance and shooting offset angle corresponding to the arthroscopic image are obtained; the movement direction indicates the direction of movement of the arthroscopic image shooting in three-dimensional space; the movement distance indicates the three-dimensional distance between the arthroscopic image shooting at the current time point and the arthroscopic image shooting at the previous time point; the shooting offset angle indicates the shooting angle of the arthroscopic image;

[0037] Superimposing the arthroscopic grayscale images according to the movement direction, shooting offset angle and movement distance to obtain an overlapping three-dimensional image;

[0038] Obtaining a joint fixation image; the joint fixation image is a three-dimensional image containing a complete joint structure stored in a database;

[0039] According to the internal structure of the joint in the joint fixed image, the joint structure in the overlapping three-dimensional image is filled to obtain a path three-dimensional image.

[0040] Optionally, obtaining a path structure diagram based on the path segment starting point, the path segment end point, the path segment, the segment direction and the segment length includes:

[0041] Adding the path segment start point and the path segment end point of the path straight line to the point set;

[0042] The line segment direction and the line segment length are used as the value of the key-value pair, the path line segment is used as the key of the key-value pair, and the path line segment is added to the edge set;

[0043] A path structure graph is constructed according to the point set and the edge set.

[0044] Optionally, obtaining a plurality of partial path skeleton point sets based on the path three-dimensional image includes:

[0045] Extracting the path area in the three-dimensional path image to obtain a path area image;

[0046] Performing skeleton extraction according to the path region image to obtain a path skeleton; the path skeleton includes a plurality of path skeleton points;

[0047] With a step size of 1, a fixed number of path skeleton points are obtained from multiple path skeleton points each time to obtain multiple partial path skeleton point sets.

[0048] Optionally, the obtaining of a plurality of path straight lines based on the partial path skeleton point set includes:

[0049] Fitting the partial path skeleton point set to obtain multiple path straight lines;

[0050] If the angle between two path straight lines corresponding to two adjacent partial path skeleton point sets is less than the angle threshold, the two path straight lines are deleted, and the values ​​in the two partial path skeleton point sets are combined to obtain a new partial skeleton point set;

[0051] The points in the new partial skeleton point set are fitted to obtain a new path straight line.

[0052] Optionally, matching the user physiological information to the path structure graph to obtain a matching graph includes:

[0053] Obtaining a key-value pair of the path structure graph; the value of the key-value pair corresponds to a set;

[0054] The user physiological information is added to a set corresponding to the value of the key-value pair corresponding to the path structure graph to obtain a matching graph.

[0055] Optionally, it also includes camera equipment, planing equipment, host controller, foot switch and LCD touch screen;

[0056] The planing device is used to plan according to the planing path;

[0057] The imaging device is used to obtain arthroscopic images at multiple time points;

[0058] The foot switch is used to control the power connection;

[0059] The host controller is used to control whether to perform an operation and to perform a video recording or planing operation;

[0060] The LCD touch screen is used for image display.

[0061] Compared with the prior art, the embodiments of the present invention achieve the following beneficial effects:

[0062] An embodiment of the present invention also provides an artificial intelligence joint planing device based on image processing, comprising a processor, wherein the processor is used to execute the following method: obtaining arthroscopic images and user physiological information at multiple time points. The user physiological information represents the physiological information corresponding to the user when collecting arthroscopic images at multiple time points. Based on the arthroscopic images at the multiple time points, the internal structure of the joint is judged, and the positions in the image are matched to obtain a three-dimensional path image. Based on the three-dimensional path image, the path is projected into a graph in a data structure to obtain a path structure graph. The path structure graph is a graph represented by a set of points and a set of edges. The user physiological information is matched to the path structure graph to obtain a matching graph. Through a Bayesian network and a temporal convolutional network, based on the matching graph and the three-dimensional path image, the path of the inserted planing device and the joint part to be planed are analyzed to obtain a planing path. The planing path represents the path completed from inserting the joint to planing the key part to be planed.

[0063] In the present invention, the three-dimensional image of the entire joint is predicted according to the arthroscopic images at multiple time points, so as to find the damaged joint position that needs to be planed. And the path can be found through the path for the controllable planing device to move. And according to the construction of the path structure diagram, the skeleton of the path is found. Because the skeleton is in the center of the path, it can maximize the satisfaction of not touching the position of the joint edge, thereby causing damage to the joint. According to the construction of key-value pairs and matching user physiological information, the characteristics of the constructed planing path on each path are saved. The path situation during the process and the change of the user's physiological information are found through the time convolution network. According to the Bayesian network, the connection mode of the network is designed under the association relationship between the path, the three-dimensional image of the path containing the internal structure of the joint and the user's physiological information, so that the detection features can be obtained more accurately, so as to reconstruct the path according to the detection features and the time features according to the position that needs to be planed. The path of the entire process from the approach to the planing of the joint part is obtained. The planing accuracy during the joint surgery is improved, and the planing path is determined according to the characteristics of the position to improve the accuracy of the planing. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] Figure 1 It is a flow chart of a method executed by a processor in an artificial intelligence joint planing device based on image processing provided by an embodiment of the present invention.

[0065] Figure 2 It is a schematic diagram of a Bayesian network in an artificial intelligence joint planing device based on image processing provided by an embodiment of the present invention.

[0066] Figure 3 It is a structural schematic diagram of a detection network in an artificial intelligence joint planing device based on image processing provided by an embodiment of the present invention.

[0067] Figure 4 It is a structural schematic diagram of the connection relationship between a detection network and a fully connected network in an artificial intelligence joint planing device based on image processing provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0068] The present invention will be described in detail below in conjunction with the accompanying drawings.

[0069] Example 1

[0070] like Figure 1 As shown, an embodiment of the present invention provides an artificial intelligence joint planing device based on image processing, and the method includes a processor.

[0071] The processor is used to execute the following method:

[0072] S101: Acquire arthroscopic images and user physiological information at multiple time points. The user physiological information refers to the physiological information of the user obtained when the arthroscopic images at multiple time points are acquired.

[0073] In this embodiment, the user's physiological information includes blood pressure and heart rate.

[0074] S102: Based on the arthroscopic images at the multiple time points, determine the internal structure of the joint, match the positions in the images, and obtain a three-dimensional path image.

[0075] S103: Based on the three-dimensional image of the path, project the path into a graph in the data structure to obtain a path structure graph. The path structure graph is a graph represented by a point set and an edge set.

[0076] S104: Matching the user's physiological information to the path structure graph to obtain a matching graph.

[0077] S105: Analyze the path of the planing device and the joint part to be planed based on the matching graph and the path three-dimensional image through the Bayesian network and the time convolution network to obtain a planing path. The planing path represents the path from the insertion joint to the planing of the key part to be planed.

[0078] Optionally, the step of analyzing the path of the planing device and the joint part to be planed based on the matching graph and the three-dimensional image of the path by using a Bayesian network and a temporal convolutional network to obtain the planing path includes:

[0079] The edge set corresponding to the matching graph is marked in the corresponding area of ​​the path three-dimensional image to obtain a marked three-dimensional image. Multiple path sets correspond to multiple marked three-dimensional images.

[0080] Since the matching graph is obtained from the path three-dimensional image, the path in the matching graph corresponds to the insertable hollow channel in the path three-dimensional image.

[0081] According to the connectivity order of the point set in the path set, the key-value pairs corresponding to the edge set in the path set are arranged to obtain a path vector.

[0082] The path vector includes values ​​in a plurality of key-value pairs, namely, line segment direction, line segment length and user physiological information.

[0083] The path vector is input into the path time convolution network to extract the features of the line segment direction, line segment length and user physiological information at different time points to obtain the time feature.

[0084] Among them, the temporal convolutional network (Temporal Convolutional Network, TCN).

[0085] Based on the multiple labeled three-dimensional images and path vectors, a Bayesian network is constructed, features are extracted, path selection is performed, and Bayesian features are obtained.

[0086] The Bayesian features and time features are input into the deconvolution network to determine the damaged joint position, reconstruct the image, and obtain the planing path.

[0087] The Bayesian feature and the time feature have the same size and are vertically stacked and then input into the deconvolution network.

[0088] The convolution kernel of the deconvolution network is larger than the size of the Bayesian feature and the time feature.

[0089] The deconvolution network outputs an image containing a planing path. The deconvolution network, the Bayesian network and the temporal convolution network are obtained by training images marked with planing paths.

[0090] Optionally, projecting the path into a graph in a data structure based on the path three-dimensional image to obtain a path structure graph includes:

[0091] Based on the path three-dimensional image, a plurality of partial path skeleton point sets are obtained. The partial path skeleton point sets are sets obtained by segmenting a plurality of path skeleton points obtained from the path three-dimensional image.

[0092] Based on the partial path skeleton point set, a plurality of path straight lines are obtained.

[0093] The intersection of the path straight lines corresponding to two adjacent partial path skeleton point sets is taken as the mutation point.

[0094] The two mutation points on the path straight line are marked according to the connection position to obtain the starting point and the end point of the path segment.

[0095] The line segment between the path line segment starting point and the path line segment end point on the path straight line is constructed as the path line segment. According to the path line segment starting point and the path line segment end point, the line segment direction and the line segment length are obtained.

[0096] If the positions of two points on the path segment are (1, 1, 1) and (3, 3, 3), the unit is millimeter. 3-1 / 3-1=1, which is equal to sine 45 degrees, so the direction of the segment is (45, 45, 45), which means that the angles of the horizontal coordinate, vertical coordinate, and vertical coordinate offset are 45 degrees, 45 degrees, and 45 degrees respectively. The length of the segment is

[0097] A path structure diagram is obtained based on the path segment starting point, the path segment end point, the path segment, the segment direction and the segment length.

[0098] Through the above method, the mutation point is found according to the skeleton. Based on the mutation point, the line segment direction and line segment length are obtained. According to multiple path skeleton points, partial path skeleton points are obtained, and a part of the points are first fitted to obtain multiple straight lines. The points with an angle difference of less than 10 degrees in two straight lines are merged and refitted.

[0099] Optionally, constructing a Bayesian network based on the multiple labeled three-dimensional images and the path vectors, extracting features, performing path selection, and obtaining Bayesian features includes:

[0100] The labeled three-dimensional image is input into a first convolutional network to obtain a first path feature. A plurality of labeled three-dimensional images correspond to a plurality of first path features.

[0101] Among them, the first convolutional network is a 3D convolutional neural network (3D Convolutional Neural Networks, 3DCNN) used for the features of three-dimensional images.

[0102] The accuracy of a second convolutional network that takes the first path feature and the line segment direction and line segment length in the path vector as input values ​​is obtained as a first accuracy.

[0103] The accuracy of the third convolutional network that obtains the user physiological information in the path vector and the first path feature as input values ​​is used as the second accuracy.

[0104] Among them, the first path features mutually influence the user physiological information in the path vector.

[0105] Among them, the third convolutional network is a fully connected neural network (Fully Connected Neural Network, referred to as FCN).

[0106] The first accuracy and the second accuracy are normalized to obtain a first normalized value and a second normalized value.

[0107] According to the first normalized value and the second normalized value as probabilities, a Bayesian network is constructed by combining the first path feature, the line segment direction and line segment length in the path vector, and the user physiological information in the path vector.

[0108] Wherein, the constructed Bayesian network is as follows: Figure 2 shown.

[0109] According to the Bayesian network, the second convolutional network and the third convolutional network are connected to obtain a detection network.

[0110] Among them, such as Bayesian network Figure 2, add a fully connected neural network (Fully Connected Neural Network, referred to as FCN), connect the second convolutional network with the fully connected network, and the third convolutional network with the fully connected neural network (Fully Connected Neural Network, referred to as FCN) to obtain a detection network.

[0111] Among them, the detection network is as follows Figure 3 shown.

[0112] The marked three-dimensional image and the path vector are input into the trained detection network according to the probability corresponding to the Bayesian network to obtain the detection feature. Multiple marked three-dimensional images correspond to multiple detection features.

[0113] Wherein, the detection network is trained by using a plurality of training data related to the joint structure.

[0114] The detection feature is input into a fully connected network, the path is detected, and the detection value is obtained. Multiple detection features correspond to multiple detection values.

[0115] Among them, the connection relationship between the detection network and the fully connected network is as follows Figure 4 shown.

[0116] Among them, the fully connected network is a fully connected neural network (Fully Connected Neural Network, referred to as FCN).

[0117] The detection feature corresponding to the detection value greater than other detection values ​​among the multiple detection values ​​is used as the Bayesian feature.

[0118] Through the above method, the probability of affecting the detection value is judged according to the association relationship between the path contained in the marked three-dimensional image and the key-value pair data contained in the path vector, the corresponding probability is replaced by the accuracy, a Bayesian network is constructed according to the probability, and a network is constructed according to the Bayesian network.

[0119] Optionally, judging the internal structure of the joint based on the arthroscopic images at the multiple time points, matching the positions in the images, and obtaining the three-dimensional path image includes:

[0120] The arthroscopic image is gray-scaled to obtain an arthroscopic gray-scale image.

[0121] Among them, Gray = (Red*0.3+Green*0.59+Blue*0.11) is used for grayscale conversion, where Red represents the value of the R channel in the arthroscopic image containing three channels of RGB. Green represents the value of the G channel in the arthroscopic image containing three channels of RGB. Blue represents the value of the B channel in the arthroscopic image containing three channels of RGB. Gray represents the grayscale value.

[0122] The movement direction, movement distance and shooting offset angle corresponding to the arthroscopic image are obtained. The movement direction indicates the direction of movement of the arthroscopic image shooting in three-dimensional space. The movement distance indicates the three-dimensional distance between the arthroscopic image shooting at the current time point and the arthroscopic image shooting at the previous time point. The shooting offset angle indicates the angle of the arthroscopic shooting.

[0123] Among them, the coordinates are constructed according to the geographical location, the ordinate is constructed in the direction of north, the abscissa is constructed in the direction of winter, the vertical coordinate is constructed in the direction perpendicular to the ground, and the position where the arthroscopy is inserted is the origin. For example, the position of the arthroscopic image at the current time point is (2, 2, 2), and the position of the arthroscopic image at the previous time point is (1, 1, 1) in millimeters. 2-1 / 2-1=1, which is equal to sine 45 degrees, so the movement direction is (45, 45, 45), indicating that the angles of the abscissa, ordinate, and vertical coordinate offset are 45 degrees, 45 degrees, and 45 degrees respectively. The movement distance is (1, 1, 1).

[0124] The arthroscopic grayscale images are superimposed according to the movement direction, shooting offset angle and movement distance to obtain an overlapping three-dimensional image.

[0125] Wherein, the joint fixation image is adjusted according to the offset angle,

[0126] A joint fixation image is obtained, wherein the joint fixation image is a three-dimensional image containing a complete joint structure stored in a database.

[0127] According to the internal structure of the joint in the joint fixed image, the joint structure in the overlapping three-dimensional image is filled to obtain a path three-dimensional image.

[0128] The positions where no joint structures exist in the joint fixed image and the joint three-dimensional image are set to -1.

[0129] Optionally, obtaining a path structure diagram based on the path segment starting point, the path segment end point, the path segment, the segment direction and the segment length includes:

[0130] The path segment start point and the path segment end point of the path straight line are added to the point set.

[0131] Among them, if the starting point of a path segment is V1 (1, 1, 1) and the end point is V2 (2, 2, 2), and the starting point of a path segment is V2 (2, 2, 2) and the end point is V3 (3, 3, 3), then the corresponding positions of V1 and V2 in the connectivity matrix are set to 1, indicating that V1 and V2 are connected, and the corresponding positions of V2 and V3 are set to 1, indicating that V2 and V3 are connected.

[0132] The line segment direction and the line segment length are used as the value of the key-value pair, the path line segment is used as the key of the key-value pair, and the path line segment is added to the edge set.

[0133] A path structure graph is constructed according to the point set and the edge set.

[0134] The path structure graph is denoted as G=(V, E), where V represents a vertex and E represents an edge. An adjacency matrix can be used to represent the connection relationship between a plurality of path segment start points and path segment end points according to the path segments.

[0135] Optionally, obtaining a plurality of partial path skeleton point sets based on the path three-dimensional image includes:

[0136] The path area in the three-dimensional path image is extracted to obtain a path area image.

[0137] Among them, the path area and the proposed orientation are set to set the value in the path area to 0 and the value in other positions to 255.

[0138] According to the path region image, skeleton extraction is performed to obtain a path skeleton, wherein the path skeleton includes a plurality of path skeleton points.

[0139] In this embodiment, the skeleton is extracted using the skeleton3d algorithm.

[0140] With a step size of 1, a fixed number of path skeleton points are obtained from multiple path skeleton points each time to obtain multiple partial path skeleton point sets.

[0141] Among them, n path skeleton points correspond to nm partial path skeleton point sets. m is a fixed number, n is the number of path skeleton points. n and m are natural numbers greater than 0.

[0142] Optionally, obtaining a plurality of partial path skeleton point sets based on the path three-dimensional image includes:

[0143] The partial path skeleton point set is fitted to obtain multiple path straight lines.

[0144] In this embodiment, the least square method is used for fitting.

[0145] If the angle between two path lines corresponding to two adjacent partial path skeleton point sets is less than the angle threshold, the two path lines are deleted, and the values ​​in the two partial path skeleton point sets are unioned to obtain a new partial skeleton point set.

[0146] In this embodiment, the angle threshold is 10 degrees.

[0147] The points in the new partial skeleton point set are fitted to obtain a new path straight line.

[0148] In this embodiment, the least square method is used for fitting.

[0149] Optionally, matching the user physiological information to the path structure graph to obtain a matching graph includes:

[0150] A key-value pair of the path structure graph is obtained, wherein the value of the key-value pair corresponds to a set.

[0151] The user's physiological information is added to a set corresponding to the value of the key-value pair corresponding to the path structure diagram to obtain a matching diagram. Optionally, it also includes a camera device, a planing device, a host controller, a foot switch and an LCD touch screen.

[0152] The planing device is used for planing according to a planing path.

[0153] The imaging device is used to obtain arthroscopic images at multiple time points.

[0154] The foot switch is used to control the power connection.

[0155] The host controller is used to control whether to operate and perform the operation of photographing or planing.

[0156] The LCD touch screen is used for image display.

Claims

1. An artificial intelligence joint planing device based on image processing, characterized in that: Including processors; The processor is used to execute the following method: Acquire arthroscopic images and user physiological information at multiple time points; the user physiological information represents the physiological information corresponding to the user when the arthroscopic images at multiple time points are acquired; Based on the arthroscopic images at the multiple time points, determine the internal structure of the joint, match the positions in the images, and obtain a three-dimensional image of the path; Based on the three-dimensional image of the path, the path is projected into a graph in the data structure to obtain a path structure graph; the path structure graph is a graph represented by a point set and an edge set; Matching the user's physiological information to the path structure graph to obtain a matching graph; Through the Bayesian network and the time convolution network, based on the matching graph and the three-dimensional image of the path, the path of the inserted planing device and the joint part to be planed are analyzed to obtain the planing path; the planing path represents the path completed from the inserted joint to the key part to be planed.

2. The artificial intelligence joint planing device based on image processing according to claim 1 is characterized in that: The method uses a Bayesian network and a time convolutional network to analyze the path of the planing device and the joint part to be planed based on the matching graph and the path three-dimensional image to obtain the planing path, including: Marking the area corresponding to the edge set corresponding to the matching graph in the path three-dimensional image to obtain a marked three-dimensional image; obtaining multiple marked three-dimensional images corresponding to multiple path sets; Arrange the key-value pairs corresponding to the edge set in the path set according to the connectivity order of the point set in the path set to obtain a path vector; Inputting the path vector into a path time convolution network, extracting features of line segment direction, line segment length and user physiological information at different time points, and obtaining time features; Based on the multiple labeled three-dimensional images and path vectors, a Bayesian network is constructed to extract features, perform path selection, and obtain Bayesian features; The Bayesian features and time features are input into the deconvolution network to determine the damaged joint position, reconstruct the image, and obtain the planing path.

3. The artificial intelligence joint planing device based on image processing according to claim 1 is characterized in that: The step of projecting the path into a graph in a data structure based on the path three-dimensional image to obtain a path structure graph includes: Based on the path three-dimensional image, a plurality of partial path skeleton point sets are obtained; the partial path skeleton point sets are sets obtained by segmenting the plurality of path skeleton points obtained from the path three-dimensional image; Based on the partial path skeleton point set, a plurality of path straight lines are obtained; The intersection of the path straight lines corresponding to two adjacent partial path skeleton point sets is taken as the mutation point; Mark the two mutation points on the path line according to the connection position to obtain the path segment start point and the path segment end point; The line segment between the starting point and the end point of the path segment on the path straight line is constructed as the path segment; and the line segment direction and the line segment length are obtained according to the starting point and the end point of the path segment; A path structure diagram is obtained based on the path segment starting point, the path segment end point, the path segment, the segment direction and the segment length.

4. The artificial intelligence joint planing device based on image processing according to claim 2 is characterized in that: The method of constructing a Bayesian network based on the multiple labeled three-dimensional images and the path vectors, extracting features, performing path selection, and obtaining Bayesian features includes: Inputting the labeled three-dimensional image into a first convolutional network to obtain a first path feature; obtaining a plurality of first path features corresponding to a plurality of labeled three-dimensional images; Obtaining the accuracy of a second convolutional network that uses the first path feature and the line segment direction and line segment length in the path vector as input values ​​as a first accuracy; obtaining the user physiological information in the path vector and the first path feature as an accuracy of a third convolutional network as input values ​​as a second accuracy; Normalizing the first accuracy and the second accuracy to obtain a first normalized value and a second normalized value; According to the first normalized value and the second normalized value as probabilities, a Bayesian network is constructed by using the first path feature, the line segment direction and line segment length in the path vector, and the user physiological information in the path vector; According to the Bayesian network, the second convolutional network and the third convolutional network are connected to obtain a detection network; Inputting the labeled three-dimensional image and the path vector into a trained detection network to obtain detection features; multiple labeled three-dimensional images correspond to multiple detection features; Input the detection feature into a fully connected network, detect the path, and obtain a detection value; multiple detection values ​​are obtained corresponding to multiple detection features; The detection feature corresponding to the detection value greater than other detection values ​​among the multiple detection values ​​is used as the Bayesian feature.

5. The artificial intelligence joint planing device based on image processing according to claim 1 is characterized in that: The method of determining the internal structure of the joint based on the arthroscopic images at the multiple time points, matching the positions in the images, and obtaining the three-dimensional path image includes: Gray-scaling the arthroscopic image to obtain an arthroscopic grayscale image; The movement direction, movement distance and shooting offset angle corresponding to the arthroscopic image are obtained; the movement direction indicates the direction of movement of the arthroscopic image shooting in three-dimensional space; the movement distance indicates the three-dimensional distance between the arthroscopic image shooting at the current time point and the arthroscopic image shooting at the previous time point; the shooting offset angle indicates the shooting angle of the arthroscopic image; Superimposing the arthroscopic grayscale images according to the movement direction, shooting offset angle and movement distance to obtain an overlapping three-dimensional image; Obtaining a joint fixation image; the joint fixation image is a three-dimensional image containing a complete joint structure stored in a database; According to the internal structure of the joint in the joint fixed image, the joint structure in the overlapping three-dimensional image is filled to obtain a path three-dimensional image.

6. The artificial intelligence joint planing device based on image processing according to claim 3 is characterized in that: The obtaining of a path structure diagram based on the path segment starting point, the path segment end point, the path segment, the segment direction and the segment length includes: Adding the path segment start point and the path segment end point of the path straight line to the point set; The line segment direction and the line segment length are used as the value of the key-value pair, the path line segment is used as the key of the key-value pair, and the path line segment is added to the edge set; A path structure graph is constructed according to the point set and the edge set.

7. The artificial intelligence joint planing device based on image processing according to claim 3 is characterized in that: The obtaining of a plurality of partial path skeleton point sets based on the path three-dimensional image comprises: Extracting the path area in the three-dimensional path image to obtain a path area image; Performing skeleton extraction according to the path region image to obtain a path skeleton; the path skeleton includes a plurality of path skeleton points; With a step size of 1, a fixed number of path skeleton points are obtained from multiple path skeleton points each time to obtain multiple partial path skeleton point sets.

8. The artificial intelligence joint planing device based on image processing according to claim 3 is characterized in that: The obtaining of a plurality of path straight lines based on the partial path skeleton point set includes: Fitting the partial path skeleton point set to obtain multiple path straight lines; If the angle between two path straight lines corresponding to two adjacent partial path skeleton point sets is less than the angle threshold, the two path straight lines are deleted, and the values ​​in the two partial path skeleton point sets are combined to obtain a new partial skeleton point set; The points in the new partial skeleton point set are fitted to obtain a new path straight line.

9. The artificial intelligence joint planing device based on image processing according to claim 1 is characterized in that: The matching of the user's physiological information to the path structure graph to obtain a matching graph includes: Obtaining a key-value pair of the path structure graph; the value of the key-value pair corresponds to a set; The user physiological information is added to a set corresponding to the value of the key-value pair corresponding to the path structure graph to obtain a matching graph.

10. The artificial intelligence joint planing device based on image processing according to claim 1 is characterized in that: It also includes camera equipment, planing equipment, host controller, foot switch and LCD touch screen; The planing device is used to plan according to the planing path; The imaging device is used to obtain arthroscopic images at multiple time points; The foot switch is used to control the power connection; The host controller is used to control whether to perform an operation and to perform a video recording or planing operation; The LCD touch screen is used for image display.

Images