Laparoscopic liver resection surgery registration method and system based on artificial intelligence algorithm

Abstract

The application provides a laparoscopic liver resection surgery registration method and system based on an artificial intelligence algorithm, and belongs to the technical field of image processing.The method carries out normalization processing on a preprocessed laparoscopic surgery image to obtain a normalized laparoscopic surgery image; an artificial intelligence algorithm is used to carry out contour extraction on the normalized laparoscopic surgery image to obtain the upper edge and the lower edge of the liver in the laparoscopic surgery image; a preoperative 3D liver model is constructed, and the upper edge and the lower edge of the liver are labeled in the constructed preoperative 3D liver model; the upper edge and the lower edge of the liver in the laparoscopic surgery image and the upper edge and the lower edge of the liver labeled in the constructed preoperative 3D liver model are subjected to registration calculation to obtain a registration result.Compared with the prior art, the application effectively solves the problem that the accuracy of laparoscopic liver resection surgery registration is not high.

Description

Technical Field

[0001] This invention belongs to the field of image processing technology, specifically relating to a registration method and system for laparoscopic liver resection surgery based on artificial intelligence algorithms. Background Technology

[0002] With the continuous development of laparoscopic technology, monocular laparoscopy is being used more and more widely in liver resection surgery. Due to the limited field of view of laparoscopic images, there will be a large deviation between the surgical field of view and the three-dimensional structure of the liver, which will bring great difficulty to the precise positioning during the operation.

[0003] Currently, most existing registration methods for laparoscopic liver resection surgery rely on manual calibration or traditional visual navigation-based image registration algorithms. These two methods not only require a lot of prior knowledge and manual operation, which leads to low accuracy in registration for laparoscopic liver resection surgery, but also make it difficult to cope with complex liver morphology and perspective changes. Summary of the Invention

[0004] The purpose of this invention is to provide a registration method and system for laparoscopic liver resection surgery based on artificial intelligence algorithms, in order to solve the problem of low accuracy in the registration of laparoscopic liver resection surgery in the prior art.

[0005] To achieve the above objectives, the present invention adopts the following technical solution:

[0006] In a first aspect, the present invention provides a registration method for laparoscopic liver resection surgery based on artificial intelligence algorithms, comprising the following steps:

[0007] Acquire images of laparoscopic surgery;

[0008] The obtained laparoscopic surgical images are preprocessed to obtain preprocessed laparoscopic surgical images. The preprocessing includes noise removal and laparoscopic surgical image enhancement.

[0009] The preprocessed laparoscopic surgical images were normalized to obtain normalized laparoscopic surgical images.

[0010] Artificial intelligence algorithms were used to extract the contours of the normalized laparoscopic surgical images to obtain the upper and lower edges of the liver in the laparoscopic surgical images;

[0011] Construct a preoperative 3D liver model and mark the upper and lower edges of the liver in the constructed preoperative 3D liver model;

[0012] The upper and lower edges of the liver in the laparoscopic surgical image were registered with the upper and lower edges of the liver marked in the preoperative 3D liver model to obtain the registration results.

[0013] A further improvement of the present invention is that the artificial intelligence algorithm is the HRNet (High-Resolution Network) model.

[0014] A further improvement of this invention lies in using the HRNet model to extract the contours of the normalized laparoscopic surgical images, obtaining the upper and lower edges of the liver in the laparoscopic surgical images, specifically including:

[0015] Training, validation, and test sets were constructed using normalized laparoscopic surgical images.

[0016] Build an HRNet model, train the HRNet model based on the built training set, evaluate the performance of the HRNet model based on the built validation set, and test the training results of the HRNet model based on the built test set to obtain a trained HRNet model.

[0017] The trained HRNet model was used to extract the contours of the normalized laparoscopic surgical images to obtain the upper and lower edges of the liver in the laparoscopic surgical images.

[0018] A further improvement of this invention is that, when training the HRNet model based on the constructed training set, the loss function used is the focus loss function, the expression of which is:

[0019]

[0020] in, For focus loss function, This is the clustering factor, used to control the degree to which the HRNet model penalizes easily classified samples. As a focus factor, it is used to reduce the weight of easily classified samples. The total number of samples, For the first Index of each sample, For the first The true label of each sample For the first The predicted probability of a sample. This represents the confidence level of the HRNet model for class 1. For the first The logarithm of the predicted probability of a sample.

[0021] A further improvement of this invention is that a channel splicing mechanism is added to improve the HRNet model when constructing the HRNet model.

[0022] A further improvement of this invention is that, after adding a channel splicing mechanism to improve the HRNet model, a convolutional block attention module is added.

[0023] A further improvement of this invention lies in the fact that the registration calculation of the upper and lower edges of the liver in the laparoscopic surgical image with the upper and lower edges of the liver marked in the constructed preoperative 3D liver model, to obtain the registration result, specifically includes:

[0024] The RANSAC-PnP algorithm was used to register the upper and lower edges of the liver in the laparoscopic surgical image with the upper and lower edges of the liver marked in the preoperative 3D liver model, and the rotation matrix, translation vector and reprojection error were obtained.

[0025] The registration result is obtained based on the rotation matrix, translation vector, and reprojection error.

[0026] Secondly, the present invention provides a laparoscopic liver resection registration system based on artificial intelligence algorithms, including a data acquisition module, a data preprocessing module, a data normalization processing module, a contour extraction module, a preoperative model construction module, and a registration module;

[0027] The data acquisition module is used to acquire laparoscopic surgical images;

[0028] The data preprocessing module is used to preprocess the obtained laparoscopic surgical images to obtain preprocessed laparoscopic surgical images. The preprocessing includes noise removal and laparoscopic surgical image enhancement.

[0029] The data normalization processing module is used to normalize the preprocessed laparoscopic surgical images to obtain normalized laparoscopic surgical images.

[0030] The contour extraction module is used to extract the contours of the normalized laparoscopic surgical image using artificial intelligence algorithms, so as to obtain the upper and lower edges of the liver in the laparoscopic surgical image.

[0031] The preoperative model building module is used to build a preoperative 3D liver model and mark the upper and lower edges of the liver in the built preoperative 3D liver model;

[0032] The registration module is used to perform registration calculations between the upper and lower edges of the liver in the laparoscopic surgical image and the upper and lower edges of the liver marked in the constructed preoperative 3D liver model, and obtain the registration result.

[0033] A further improvement of this invention is that the artificial intelligence algorithm is the HRNet model.

[0034] A further improvement of this invention is that the contour extraction module uses the HRNet model to extract the contours of the normalized laparoscopic surgical image, obtaining the upper and lower edges of the liver in the laparoscopic surgical image, specifically including:

[0035] Training, validation, and test sets were constructed using normalized laparoscopic surgical images.

[0036] Build an HRNet model, train the HRNet model based on the built training set, evaluate the performance of the HRNet model based on the built validation set, and test the training results of the HRNet model based on the built test set to obtain a trained HRNet model.

[0037] The trained HRNet model was used to extract the contours of the normalized laparoscopic surgical images to obtain the upper and lower edges of the liver in the laparoscopic surgical images.

[0038] Compared with the prior art, the present invention has the following beneficial effects:

[0039] Compared with existing registration methods for laparoscopic liver resection, this invention, on the one hand, employs artificial intelligence algorithms to extract contours from normalized laparoscopic surgical images, obtaining the upper and lower edges of the liver in the laparoscopic surgical images. The upper and lower edges of the liver obtained through contour extraction possess specific local features, which can more accurately match different images (laparoscopic surgical images) during subsequent registration, thereby improving registration accuracy. On the other hand, this invention performs registration calculations by matching the upper and lower edges of the liver in the laparoscopic surgical images with the upper and lower edges of the liver marked in the constructed preoperative 3D liver model. This achieves precise alignment between the intraoperative image (laparoscopic surgical image) and the 3D model (preoperative 3D liver model), reducing manual intervention and further improving registration accuracy. This effectively solves the problem of low registration accuracy in existing technologies for laparoscopic liver resection. Attached Figure Description

[0040] Figure 1 This is a flowchart of the registration method for laparoscopic liver resection surgery based on artificial intelligence algorithms according to the present invention;

[0041] Figure 2 This is a schematic diagram of the laparoscopic liver resection registration system based on artificial intelligence algorithms of the present invention;

[0042] Figure 3 The present invention provides laparoscopic surgical images based on artificial intelligence algorithms and the results of contour extraction from laparoscopic surgical images.

[0043] Figure 4 This is a structural diagram of the convolutional block attention module of the present invention;

[0044] Figure 5 This is a structural diagram of the HRNet model ultimately used in this invention;

[0045] Figure 6 This is a schematic diagram of the preoperative 3D liver model constructed according to the present invention. Detailed Implementation

[0046] To further understand the content of this invention, the invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments are merely illustrative and not limiting of the invention.

[0047] This invention proposes a registration method for laparoscopic liver resection surgery based on artificial intelligence algorithms. The method performs registration calculations between the upper and lower edges of the liver in the laparoscopic surgical image and the upper and lower edges of the liver marked in the constructed preoperative 3D liver model, obtaining the registration result. Compared with existing technologies, this invention effectively solves the problem of low accuracy in registration for laparoscopic liver resection surgery in existing technologies.

[0048] Example 1:

[0049] The flowchart of the registration method for laparoscopic liver resection surgery based on artificial intelligence algorithm of this invention is as follows: Figure 1 As shown, the laparoscopic liver resection registration method based on artificial intelligence algorithms of the present invention includes the following steps:

[0050] S1. Acquire images of laparoscopic surgery;

[0051] S2. The obtained laparoscopic surgical images are preprocessed to obtain preprocessed laparoscopic surgical images. The preprocessing includes noise removal and laparoscopic surgical image enhancement.

[0052] S3. Normalize the preprocessed laparoscopic surgical images to obtain normalized laparoscopic surgical images;

[0053] S4. Artificial intelligence algorithms are used to extract the contours of the normalized laparoscopic surgical images to obtain the upper and lower edges of the liver in the laparoscopic surgical images;

[0054] S5. Construct a preoperative 3D liver model and mark the upper and lower edges of the liver in the constructed preoperative 3D liver model;

[0055] S6. Perform registration calculations between the upper and lower edges of the liver in the laparoscopic surgical image and the upper and lower edges of the liver marked in the constructed preoperative 3D liver model to obtain the registration results.

[0056] Example 2:

[0057] A schematic diagram of the laparoscopic liver resection registration system based on artificial intelligence algorithms of this invention is shown below. Figure 2 As shown, the laparoscopic liver resection registration system based on artificial intelligence algorithms of the present invention includes a data acquisition module, a data preprocessing module, a data normalization processing module, a contour extraction module, a preoperative model construction module, and a registration module.

[0058] The data acquisition module is used to acquire images of laparoscopic surgery.

[0059] The data preprocessing module is used to preprocess the obtained laparoscopic surgical images to obtain preprocessed laparoscopic surgical images. The preprocessing includes noise removal and laparoscopic surgical image enhancement.

[0060] The data normalization module is used to normalize the preprocessed laparoscopic surgical images to obtain normalized laparoscopic surgical images.

[0061] The contour extraction module is used to extract the contours of the normalized laparoscopic surgical images using artificial intelligence algorithms, thereby obtaining the upper and lower edges of the liver in the laparoscopic surgical images.

[0062] The preoperative model building module is used to build a preoperative 3D liver model and mark the upper and lower edges of the liver in the built preoperative 3D liver model.

[0063] The registration module is used to perform registration calculations between the upper and lower edges of the liver in the laparoscopic surgical image and the upper and lower edges of the liver marked in the constructed preoperative 3D liver model, and obtain the registration results.

[0064] Example 3:

[0065] The registration method for laparoscopic liver resection surgery based on artificial intelligence algorithms of the present invention includes the following steps:

[0066] S1. Acquire images of laparoscopic surgery.

[0067] S2. Preprocess the obtained laparoscopic surgical images to obtain preprocessed laparoscopic surgical images. Preprocessing includes noise removal and laparoscopic surgical image enhancement.

[0068] This step specifically employs Gaussian filtering or bilateral filtering to remove noise, and enhances the laparoscopic surgical images by adjusting the brightness and saturation of the HSV (Hue, Saturation, Value) or CIELAB (International Commission on Illumination LAB Color) space.

[0069] S3. Normalize the preprocessed laparoscopic surgical images to obtain normalized laparoscopic surgical images.

[0070] The formula for normalization in this step is:

[0071]

[0072] in, These are pre-processed images of laparoscopic surgery. These are normalized images of laparoscopic surgery. This represents the minimum value of the preprocessed laparoscopic surgical image. The value is usually 0. This represents the maximum value of the preprocessed laparoscopic surgical image. The value is usually set to 255.

[0073] S4. Artificial intelligence algorithms are used to extract the contours of the normalized laparoscopic surgical images to obtain the upper and lower edges of the liver in the laparoscopic surgical images.

[0074] This step specifically uses the HRNet model to extract the contours of the normalized laparoscopic surgical images, obtaining the upper and lower edges of the liver in the laparoscopic surgical images.

[0075] This step uses the HRNet model to extract the contours of the normalized laparoscopic surgical images, obtaining the upper and lower edges of the liver in the laparoscopic surgical images, specifically including:

[0076] A. Construct training, validation, and test sets using normalized laparoscopic surgical images;

[0077] B. Construct the HRNet model, train the HRNet model based on the constructed training set, evaluate the performance of the HRNet model based on the constructed validation set, and test the training results of the HRNet model based on the constructed test set to obtain the trained HRNet model.

[0078] C. The trained HRNet model is used to extract the contours of the normalized laparoscopic surgical images to obtain the upper and lower edges of the liver in the laparoscopic surgical images.

[0079] Step A specifically includes 921 training images (training set), 122 validation images (validation set), and 109 test images (test set). Each image has a pixel size of 1920×1080, and the size of each image is uniformly fixed at 1024×1024 during training. The laparoscopic surgery images and the results of contour extraction from the laparoscopic surgery images of this invention are as follows... Figure 3 As shown, Figure 3 Image 'a' represents a laparoscopic surgery image. Figure 3 b represents the result after contour extraction from the laparoscopic surgical image. Figure 3 In image b, the green portion represents the upper edge (also called the upper margin) of the liver in a laparoscopic surgical image. Figure 3 In diagram b, the red portion represents the lower edge (also called the lower margin) of the liver in the laparoscopic surgical image, and the blue portion represents the resected falciform ligament. Figure 3 The black area in b is the background.

[0080] In step B, when constructing the HRNet model, a channel splicing mechanism is added to improve the HRNet model.

[0081] In step B, the performance metrics for evaluating the HRNet model based on the constructed validation set include precision, recall, F1 score, and mean intersection over union (mIoU).

[0082] This embodiment compares the evaluation metrics of the four existing models and the HRNet model. The comparison results are shown in Table 1.

[0083] Table 1 Comparison of Evaluation Indicator Results

[0084]

[0085] The data in Table 1 shows that the HRNet model outperforms the other four models in all four metrics. Therefore, in this embodiment, the HRNet model is selected to extract the contours of the normalized laparoscopic surgical images to obtain the upper and lower edges of the liver in the laparoscopic surgical images.

[0086] In step B, when training the HRNet model based on the constructed training set, the loss function used is the focus loss function, and the expression for the focus loss function is:

[0087]

[0088] in, For focus loss function, This is the clustering factor, used to control the degree to which the HRNet model penalizes easily classified samples. The value is usually 2. As a focus factor, it is used to reduce the weight of easily classified samples. The total number of samples, For the first Index of each sample, For the first The true label of each sample For the first The predicted probability of a sample. This represents the confidence level of the HRNet model for class 1. For the first The logarithm of the predicted probability of a sample.

[0089] After improving the HRNet model by adding a channel concatenation mechanism in step B, a Convolutional Block Attention Module (CBAM) is added. The CBAM is explained in detail below:

[0090] CBAM is a lightweight and efficient attention mechanism commonly used in Convolutional Neural Networks (CNNs) to enhance feature representation capabilities. By introducing an attention mechanism, CBAM enables CNNs to focus more on key spatial and channel information.

[0091] The core idea of ​​CBAM is to model attention in both the channel and spatial dimensions of the feature map. CBAM mainly consists of a Channel Attention Module and a Spatial Attention Module. Channel attention allows the convolutional neural network to focus on which channels contain more important information. Channel attention uses global average pooling and global max pooling to extract features in the channel dimension, representing different statistical information of the feature map. The structure diagram of the convolutional block attention module is shown below. Figure 4 As shown.

[0092] This embodiment compares the evaluation metrics results after the HRNet model is optimized, and the comparison results are shown in Table 2.

[0093] Table 2 Comparison of Evaluation Indicator Results

[0094]

[0095] In Table 2, optimization 1 uses the focus loss function when training the HRNet model based on the constructed training set. Optimization 2 adds a channel concatenation mechanism to improve the HRNet model during construction. Optimization 3 adds a convolutional block attention module after improving the HRNet model with the channel concatenation mechanism. When only optimizations 2 and 3 are used, the cross-entropy loss function is used when training the HRNet model based on the constructed training set.

[0096] As can be seen from Table 2, the evaluation metrics (precision, recall, F1 score, and mean intersection-over-union ratio) of the optimized HRNet model are generally improved compared to the evaluation metrics (precision, recall, F1 score, and mean intersection-over-union ratio) of the HRNet model in Table 1.

[0097] The structure diagram of the HRNet model finally adopted in this invention is as follows: Figure 5 As shown below, the process of obtaining the upper and lower edges of the liver in laparoscopic surgical images is explained in detail:

[0098] First, convolution and downsampling operations are performed on the normalized laparoscopic surgical image to obtain four parallel branches, each with a different feature map resolution. Then, in the first three branches, channel stitching (also called adding a channel stitching mechanism) is performed on feature maps of the same resolution to obtain three feature maps of different resolutions. Finally, the feature maps of the three different resolutions and the feature map of the fourth branch are processed simultaneously using a convolutional block attention module (also called a CBAM attention module). After processing the feature maps of the three different resolutions and the feature map of the fourth branch simultaneously using the convolutional block attention module, an upsampling operation is performed to obtain the upper and lower edges of the liver in the laparoscopic surgical image.

[0099] S5. Construct a preoperative 3D liver model and mark the upper and lower edges of the liver in the constructed preoperative 3D liver model.

[0100] A preoperative 3D liver model was constructed, and the upper and lower edges of the liver were marked on the model. A schematic diagram of the constructed preoperative 3D liver model is shown below. Figure 6 As shown.

[0101] S6. Perform registration calculations between the upper and lower edges of the liver in the laparoscopic surgical image and the upper and lower edges of the liver marked in the constructed preoperative 3D liver model to obtain the registration results.

[0102] The upper and lower edges of the liver in the laparoscopic surgical image are registered with the upper and lower edges of the liver marked in the constructed preoperative 3D liver model. The registration results are obtained, including:

[0103] a. The RANSAC-PnP algorithm was used to register the upper and lower edges of the liver in the laparoscopic surgical image with the upper and lower edges of the liver marked in the preoperative 3D liver model, and the rotation matrix, translation vector and reprojection error were obtained.

[0104] The specific process of obtaining the rotation matrix in this step is explained below:

[0105] First, define three unit vectors. , and , , and The calculation formula is:

[0106]

[0107]

[0108]

[0109] in, , and Let x be a unit vector, representing the direction to the right of the camera. Indicates the direction of the camera upwards. Indicates the direction of the camera's line of sight (from the focal point to the camera). Let be the direction vector on the camera. For camera position, The focal point.

[0110] Then through unit vectors , and The final rotation matrix is:

[0111]

[0112] in, Let be a rotation matrix. for vector Coordinate values for vector Coordinate values for vector Coordinate values for vector Coordinate values for vector Coordinate values for vector Coordinate values for vector Coordinate values for vector Coordinate values for vector Coordinate values.

[0113] The formula for calculating the translation vector in this step is:

[0114]

[0115] in, It is a translation vector. Let be a rotation matrix. This indicates the camera position.

[0116] Among them, the above-mentioned camera positions Focus position and the direction vector on the camera All were obtained using the vtkcamera rendering library in the vtk data visualization development kit.

[0117] The formula for calculating the reprojection error in this step is:

[0118]

[0119] in, For reprojection error, The number of points participating in the registration. For the first Reprojection error per pixel For a pixel in a laparoscopic surgical image, For the rotation matrix The calculated pixel coordinates, The row coordinates of the pixel. express and Euclidean distance.

[0120] b. Based on the rotation matrix, translation vector, and reprojection error, the registration result is obtained.

[0121] In this step, when the rotation matrix, translation vector, and reprojection error all meet the set conditions, it means that the upper and lower edges of the liver in the laparoscopic surgical image can be registered with the upper and lower edges of the liver marked in the constructed preoperative 3D liver model.

[0122] Specifically, when the rotation matrix and translation vector can make the upper and lower edges of the liver in the laparoscopic surgical image coincide with the upper and lower edges of the liver marked in the constructed preoperative 3D liver model within a set range (the set range is adjusted according to actual needs), it means that the upper and lower edges of the liver in the laparoscopic surgical image can be registered with the upper and lower edges of the liver marked in the constructed preoperative 3D liver model.

[0123] Specifically, when the reprojection error meets the set error threshold (the error threshold can be adjusted according to actual needs in this step, and is generally set to 4-5mm), it means that the upper and lower edges of the liver in the laparoscopic surgical image can be registered with the upper and lower edges of the liver marked in the constructed preoperative 3D liver model.

[0124] The registration result is finally obtained based on the registration calculation process.

[0125] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A registration method for laparoscopic liver resection surgery based on artificial intelligence algorithms, characterized in that, Includes the following steps: Acquire images of laparoscopic surgery; The obtained laparoscopic surgical images are preprocessed to obtain preprocessed laparoscopic surgical images. The preprocessing includes noise removal and laparoscopic surgical image enhancement. The preprocessed laparoscopic surgical images were normalized to obtain normalized laparoscopic surgical images. Artificial intelligence algorithms were used to extract the contours of the normalized laparoscopic surgical images to obtain the upper and lower edges of the liver in the laparoscopic surgical images; Construct a preoperative 3D liver model and mark the upper and lower edges of the liver in the constructed preoperative 3D liver model; The upper and lower edges of the liver in the laparoscopic surgical image were registered with the upper and lower edges of the liver marked in the preoperative 3D liver model to obtain the registration results.

2. The registration method for laparoscopic liver resection surgery based on artificial intelligence algorithm according to claim 1, characterized in that, The artificial intelligence algorithm is the HRNet model.

3. The registration method for laparoscopic liver resection surgery based on artificial intelligence algorithms according to claim 2, characterized in that, The HRNet model was used to extract the contours of the normalized laparoscopic surgical images, obtaining the upper and lower edges of the liver in the laparoscopic surgical images, specifically including: Training, validation, and test sets were constructed using normalized laparoscopic surgical images. Build an HRNet model, train the HRNet model based on the built training set, evaluate the performance of the HRNet model based on the built validation set, and test the training results of the HRNet model based on the built test set to obtain a trained HRNet model. The trained HRNet model was used to extract the contours of the normalized laparoscopic surgical images to obtain the upper and lower edges of the liver in the laparoscopic surgical images.

4. The registration method for laparoscopic liver resection surgery based on artificial intelligence algorithm according to claim 3, characterized in that, When training the HRNet model using the constructed training set, the focus loss function is used, and its expression is as follows: in, For focus loss function, This is the clustering factor, used to control the degree to which the HRNet model penalizes easily classified samples. As a focus factor, it is used to reduce the weight of easily classified samples. The total number of samples, For the first Index of each sample, For the first The true label of each sample For the first The predicted probability of a sample. This represents the confidence level of the HRNet model for class 1. For the first The logarithm of the predicted probability of a sample.

5. The registration method for laparoscopic liver resection surgery based on artificial intelligence algorithms according to claim 3, characterized in that, When building the HRNet model, a channel splicing mechanism is added to improve the HRNet model.

6. The registration method for laparoscopic liver resection surgery based on artificial intelligence algorithms according to claim 5, characterized in that, After improving the HRNet model by adding a channel splicing mechanism, a convolutional block attention module was added.

7. The registration method for laparoscopic liver resection surgery based on artificial intelligence algorithm according to claim 1, characterized in that, The registration calculation is performed between the upper and lower edges of the liver in the laparoscopic surgical image and the upper and lower edges of the liver marked in the constructed preoperative 3D liver model to obtain the registration result, which specifically includes: The RANSAC-PnP algorithm was used to register the upper and lower edges of the liver in the laparoscopic surgical image with the upper and lower edges of the liver marked in the preoperative 3D liver model, and the rotation matrix, translation vector and reprojection error were obtained. The registration result is obtained based on the rotation matrix, translation vector, and reprojection error.

8. A registration system for laparoscopic liver resection surgery based on artificial intelligence algorithms, characterized in that, It includes a data acquisition module, a data preprocessing module, a data normalization module, a contour extraction module, a preoperative model building module, and a registration module; The data acquisition module is used to acquire laparoscopic surgical images; The data preprocessing module is used to preprocess the obtained laparoscopic surgical images to obtain preprocessed laparoscopic surgical images. The preprocessing includes noise removal and laparoscopic surgical image enhancement. The data normalization processing module is used to normalize the preprocessed laparoscopic surgical images to obtain normalized laparoscopic surgical images. The contour extraction module is used to extract the contours of the normalized laparoscopic surgical image using artificial intelligence algorithms, so as to obtain the upper and lower edges of the liver in the laparoscopic surgical image. The preoperative model building module is used to build a preoperative 3D liver model and mark the upper and lower edges of the liver in the built preoperative 3D liver model; The registration module is used to perform registration calculations between the upper and lower edges of the liver in the laparoscopic surgical image and the upper and lower edges of the liver marked in the constructed preoperative 3D liver model, and obtain the registration result.

9. The laparoscopic liver resection registration system based on artificial intelligence algorithms according to claim 8, characterized in that, The artificial intelligence algorithm is the HRNet model.

10. The laparoscopic liver resection registration system based on artificial intelligence algorithms according to claim 9, characterized in that, The contour extraction module uses the HRNet model to extract the contours of the normalized laparoscopic surgical images, obtaining the upper and lower edges of the liver in the laparoscopic surgical images, specifically including: Training, validation, and test sets were constructed using normalized laparoscopic surgical images. Build an HRNet model, train the HRNet model based on the built training set, evaluate the performance of the HRNet model based on the built validation set, and test the training results of the HRNet model based on the built test set to obtain a trained HRNet model. The trained HRNet model was used to extract the contours of the normalized laparoscopic surgical images to obtain the upper and lower edges of the liver in the laparoscopic surgical images.

Images