An intelligent nursing edge positioning and adjustment system based on image recognition

Abstract

The present invention relates to the field of image recognition, and in particular to an intelligent nursing edge positioning and adjustment system based on image recognition, comprising an image acquisition module, an intelligent image fusion module, an intelligent image segmentation module, an edge positioning adjustment module and a user interface module; the present invention creatively adopts a multi-scale feature fusion method based on multi-dimensional dynamic convolution, extracts detailed features of lesion images through multi-dimensional dynamic convolution, enhances the connection of feature context information through multiple channel attention, enhances visual authenticity and perceived details, learns the structural texture of the lesion, and improves the accuracy of edge positioning; the present invention creatively adopts a dual encoder network based on multi-dimensional information extraction, captures the correlation information between the space and channels of the fused image through convolutional attention encoding, captures the global and local information of the fused image through the encoder of Swin-Unet, fuses and decodes the two types of information, accurately segments the lesion site and uses it for edge positioning.

Description

Technical Field

[0001] The invention belongs to the field of image recognition, and in particular is an intelligent nursing edge positioning and adjustment system based on image recognition. Background Art

[0002] Nursing edge positioning is one of the key steps in medical care. It refers to the positioning and fixing of nursing supplies in the medical field. It aims to ensure the correct position of nursing supplies on the patient and effectively fix them to ensure the treatment effect and patient comfort. However, the existing nursing edge positioning is usually manual, which is inefficient and has low positioning accuracy. The existing auxiliary positioning method is usually simple medical image segmentation. Single-modal medical image segmentation cannot provide more comprehensive positioning information for edge positioning. Simple multimodal image fusion cannot achieve a balance between global consistency and local accuracy, and the positioning accuracy is low. Summary of the invention

[0003] In view of the above problems, in order to overcome the defects of the prior art, the present invention creatively provides an intelligent nursing edge positioning and adjustment system based on image recognition. In view of the low efficiency and low positioning accuracy of the manual edge positioning method, the present invention adopts an image recognition method based on computer vision to increase the accuracy and efficiency of positioning by identifying and segmenting the patient's lesion image; in view of the problem that the existing positioning method cannot provide more comprehensive positioning information for edge positioning and cannot achieve a balance between global consistency and local accuracy, the present invention creatively adopts a multi-scale feature fusion method based on multi-dimensional dynamic convolution. The detailed features of the patient's lesion image are extracted by convolution, and the features are activated and fused through multiple channel attentions, so as to enhance the connection of feature context information, enhance the visual reality and perceptual details of the fused image, and learn the structural texture of the lesion to improve the accuracy of edge positioning. The present invention creatively adopts a dual encoder network based on multi-dimensional information extraction, captures the correlation information between the space and channels of the fused image through convolutional attention coding, captures the global and local information of the fused image through the Swin-Unet encoder, and connects and decodes the two types of information after transposed convolution, so as to obtain a more accurate lesion location for precise edge positioning.

[0004] The present invention provides an intelligent nursing edge positioning and adjustment system based on image recognition, comprising an image acquisition module, an intelligent image fusion module, an intelligent image segmentation module, an edge positioning adjustment module and a user interface module;

[0005] The image acquisition module uses a color camera to capture an RGB image of the patient's lesion, uses a depth camera to capture a depth image of the patient's lesion, and sends the RGB image and the depth image to the intelligent image fusion module;

[0006] The intelligent image fusion module uses a multi-scale feature fusion method based on multi-dimensional dynamic convolution to fuse the RGB image and the depth image, obtains a grayscale fusion image and sends it to the intelligent image segmentation module;

[0007] The intelligent image segmentation module uses a dual encoder network based on multi-dimensional information extraction, captures the spatial and channel correlation information of the fused image through convolutional attention coding, captures the global and local information of the fused image through the Swin-Unet encoder, and performs convolution and feature selection on the two types of information and then performs concatenation decoding to obtain a segmented image, and finally sends the segmented image to the edge positioning adjustment module;

[0008] The edge positioning adjustment module determines the edge position and adjustment direction according to the segmented image to obtain the edge positioning adjustment result;

[0009] The user interface module allows medical staff to view the results of the edge positioning adjustment, and the medical staff can accurately edge the patient's lesion based on the results. It also allows technical personnel to adjust parameters of all modules and view information.

[0010] Furthermore, in the intelligent image fusion module, a multi-scale feature fusion method based on multi-dimensional dynamic convolution specifically includes the following steps:

[0011] Step S1: Image conversion, converting the RGB image to the YCbCr space to obtain the brightness component image Y, the color difference component image Cb and the color difference component image Cr of the RGB image, and converting the depth image in a depth-to-brightness conversion manner to obtain the brightness image D of the depth image;

[0012] Step S2: Multi-dimensional dynamic convolution, perform multi-dimensional attention dynamic convolution on the brightness component map Y and the brightness map D respectively, and obtain the brightness component convolution map and the brightness convolution map respectively:

[0013] ;

[0014] In the formula, represents the Gaussian error linear unit activation function, represents the normalization operation, It stands for multi-dimensional attention dynamic convolution, which includes four dimensions of attention dynamic convolution, namely spatial dimension, input channel dimension, output channel dimension and convolution kernel number dimension. The spatial dimension corresponds to the spatial position of each convolution kernel, the input channel dimension corresponds to the channel dimension of the input data, the output channel dimension corresponds to the channel dimension of the output data, and the convolution kernel number dimension corresponds to the weights between different convolution kernels. Represents the input image, namely the brightness component image Y and the brightness image D, Represents the output image, i.e., the brightness component convolution map and the brightness convolution map;

[0015] Step S3: multi-modal feature activation, performing multi-modal feature activation on the brightness component convolution map and the brightness convolution map respectively, to obtain brightness component activation features and brightness activation features respectively;

[0016] Step S4: image fusion, fusing and activating the brightness component activation feature and the brightness activation feature to obtain a brightness fusion image;

[0017] Step S5: grayscale image conversion, converting the brightness fusion image in a brightness-to-grayscale conversion manner to obtain a grayscale fusion image, whose dimension is (W, H, C).

[0018] Further, step S3 specifically includes the following steps:

[0019] Step S31: The brightness component convolution map and the brightness convolution map are collectively referred to as the input image. The input image is subjected to LN normalization processing, and the window-based multi-head self-attention processing and channel attention module processing are respectively performed. The two processing results are added to the original input image to obtain the primary attention feature:

[0020] ;

[0021] In the formula, Represents the input image, namely the brightness component convolution map and the brightness convolution map, represents LN normalization, represents window-based multi-head self-attention processing, represents the channel attention module processing, represents primary attention features;

[0022] Step S32: Perform LN normalization on the primary attention features and perform multi-layer perceptron processing, and add the processing results to the primary attention features to obtain preliminary activation features:

[0023] ;

[0024] In the formula, represents multi-layer perceptron processing, represents the initial activation feature;

[0025] Step S33: Perform LN normalization on the preliminary activation features and perform overlapping attention module processing, and add the processing result to the preliminary activation features to obtain secondary attention features:

[0026] ;

[0027] In the formula, represents overlapping cross-attention module processing, represents secondary attention features;

[0028] Step S34: perform LN normalization on the secondary attention feature and perform multi-layer perceptron processing, and add the processing result to the secondary attention feature to obtain the secondary activation feature:

[0029] ;

[0030] In the formula, represents multi-layer perceptron processing, Represents the secondary activation features, namely the brightness component activation feature and the brightness activation feature.

[0031] Furthermore, in the intelligent image segmentation module, a dual encoder network based on multi-dimensional information extraction specifically includes the following processing steps:

[0032] Step Q1: CBAM encoding, use the encoder CBAM based on the convolutional block attention mechanism to encode the grayscale fusion image in n layers to obtain the convolutional coding features;

[0033] Step Q2: Swin encoding, use the Swin-Unet encoder to encode the grayscale fusion image in n layers to obtain the variable voltage coding feature;

[0034] Step Q3: Feature fusion decoding, after transposing the convolutional coding features, concatenate them with the transformer coding features, and perform 1*1 convolution and batch normalization to obtain the primary decoding result;

[0035] Step Q4: Feature selection, use the channel attention SE module (Squeeze-and-Excitation module) to select features from the primary decoding results to obtain primary selected features;

[0036] Step Q5: Multi-layer decoding, perform n-1 layer upsampling on the primary selected features to obtain the final decoded features, and use the activation function to activate the final decoded features to obtain the segmented image.

[0037] The beneficial results achieved by the present invention using the above scheme are as follows:

[0038] (1) To address the problem of low efficiency and low positioning accuracy of manual edge positioning, the present invention adopts an image recognition method based on computer vision to increase the accuracy and efficiency of positioning by identifying and segmenting the patient's lesion image;

[0039] (2) In view of the problem that the existing positioning methods cannot provide more comprehensive positioning information for edge positioning and cannot strike a balance between global consistency and local accuracy, the present invention creatively adopts a multi-scale feature fusion method based on multi-dimensional dynamic convolution, extracts detailed features of the patient's lesion image through multi-dimensional dynamic convolution, and activates and fuses the features through multiple channel attention, enhances the connection of feature context information, enhances the visual authenticity and perceptual details of the fused image, and simultaneously learns the structural texture of the lesion to improve the accuracy of edge positioning;

[0040] (3) The present invention creatively adopts a dual encoder network based on multi-dimensional information extraction, captures the spatial and channel correlation information of the fused image through convolutional attention coding, captures the global and local information of the fused image through the Swin-Unet encoder, and concatenates and decodes the two types of information after transposition convolution, so as to obtain a more accurate lesion location and use it for precise edge positioning. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 A module diagram of an intelligent nursing edge positioning and adjustment system based on image recognition provided by the present invention;

[0042] Figure 2 It is a flowchart of a multi-scale feature fusion method based on multi-dimensional dynamic convolution;

[0043] Figure 3 Flowchart of a dual encoder network based on multi-dimensional information extraction.

[0044] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0045] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0046] Example 1, see Figure 1 , the present invention provides an intelligent nursing edge positioning and adjustment system based on image recognition, including an image acquisition module, an intelligent image fusion module, an intelligent image segmentation module, an edge positioning adjustment module and a user interface module;

[0047] The image acquisition module uses a color camera to capture an RGB image of the patient's lesion, uses a depth camera to capture a depth image of the patient's lesion, and sends the RGB image and the depth image to the intelligent image fusion module;

[0048] The intelligent image fusion module uses a multi-scale feature fusion method based on multi-dimensional dynamic convolution to fuse the RGB image and the depth image, obtains a grayscale fusion image and sends it to the intelligent image segmentation module;

[0049] The intelligent image segmentation module uses a dual encoder network based on multi-dimensional information extraction, captures the spatial and channel correlation information of the fused image through convolutional attention coding, captures the global and local information of the fused image through the Swin-Unet encoder, and performs convolution and feature selection on the two types of information and then performs concatenation decoding to obtain a segmented image, and finally sends the segmented image to the edge positioning adjustment module;

[0050] The edge positioning adjustment module determines the edge position and adjustment direction according to the segmented image to obtain the edge positioning adjustment result;

[0051] The user interface module allows medical staff to view the results of the edge positioning adjustment, and the medical staff can accurately edge the patient's lesion based on the results. It also allows technical personnel to adjust parameters of all modules and view information.

[0052] By performing the above operations, in order to solve the problem of low efficiency and low positioning accuracy of the manual edge positioning method, the present invention adopts an image recognition method based on computer vision to increase the accuracy and efficiency of positioning by identifying and segmenting the patient's lesion image.

[0053] Example 2, see Figure 2 Based on the above embodiment, this embodiment provides a multi-scale feature fusion method based on multi-dimensional dynamic convolution in an intelligent image fusion module, which specifically includes the following steps:

[0054] Step S1: Image conversion, converting the RGB image to the YCbCr space to obtain the brightness component image Y, the color difference component image Cb and the color difference component image Cr of the RGB image, and converting the depth image in a depth-to-brightness conversion manner to obtain the brightness image D of the depth image;

[0055] Step S2: Multi-dimensional dynamic convolution, perform multi-dimensional attention dynamic convolution on the brightness component map Y and the brightness map D respectively, and obtain the brightness component convolution map and the brightness convolution map respectively:

[0056] ;

[0057] In the formula, represents the Gaussian error linear unit activation function, represents the normalization operation, It stands for multi-dimensional attention dynamic convolution, which includes four dimensions of attention dynamic convolution, namely spatial dimension, input channel dimension, output channel dimension and convolution kernel number dimension. The spatial dimension corresponds to the spatial position of each convolution kernel, the input channel dimension corresponds to the channel dimension of the input data, the output channel dimension corresponds to the channel dimension of the output data, and the convolution kernel number dimension corresponds to the weights between different convolution kernels. Represents the input image, namely the brightness component image Y and the brightness image D, Represents the output image, i.e., the brightness component convolution map and the brightness convolution map;

[0058] Step S3: multi-modal feature activation, performing multi-modal feature activation on the brightness component convolution map and the brightness convolution map respectively, to obtain brightness component activation features and brightness activation features respectively;

[0059] Step S4: image fusion, fusing and activating the brightness component activation feature and the brightness activation feature to obtain a brightness fusion image;

[0060] Step S5: grayscale image conversion, converting the brightness fusion image in a brightness-to-grayscale conversion manner to obtain a grayscale fusion image, whose dimension is (W, H, C).

[0061] Embodiment 3: This embodiment is based on the above embodiment, and step S3 specifically includes the following steps:

[0062] Step S31: The brightness component convolution map and the brightness convolution map are collectively referred to as the input image. The input image is subjected to LN normalization processing, and the window-based multi-head self-attention processing and channel attention module processing are respectively performed. The two processing results are added to the original input image to obtain the primary attention feature:

[0063] ;

[0064] In the formula, Represents the input image, namely the brightness component convolution map and the brightness convolution map, represents LN normalization, represents window-based multi-head self-attention processing, represents the channel attention module processing, represents primary attention features;

[0065] Step S32: Perform LN normalization on the primary attention features and perform multi-layer perceptron processing, and add the processing results to the primary attention features to obtain preliminary activation features:

[0066] ;

[0067] In the formula, represents multi-layer perceptron processing, represents the initial activation feature;

[0068] Step S33: Perform LN normalization on the preliminary activation features and perform overlapping attention module processing, and add the processing result to the preliminary activation features to obtain secondary attention features:

[0069] ;

[0070] In the formula, represents overlapping cross-attention module processing, represents secondary attention features;

[0071] Step S34: perform LN normalization on the secondary attention feature and perform multi-layer perceptron processing, and add the processing result to the secondary attention feature to obtain the secondary activation feature:

[0072] ;

[0073] In the formula, represents multi-layer perceptron processing, Represents the secondary activation features, namely the brightness component activation feature and the brightness activation feature.

[0074] By performing the above operations, in order to address the problem that existing positioning methods cannot provide more comprehensive positioning information for edge positioning and cannot achieve a balance between global consistency and local accuracy, the present invention creatively adopts a multi-scale feature fusion method based on multi-dimensional dynamic convolution. The detailed features of the patient's lesion image are extracted through multi-dimensional dynamic convolution, and the features are activated and fused through multiple channel attentions to enhance the connection of feature context information, enhance the visual authenticity and perceptual details of the fused image, and simultaneously learn the structural texture of the lesion to improve the accuracy of edge positioning.

[0075] Example 4, see Figure 3 Based on the above embodiment, in the intelligent image segmentation module, this embodiment includes a dual encoder network based on multi-dimensional information extraction, which specifically includes the following processing steps:

[0076] Step Q1: CBAM encoding, use the encoder CBAM based on the convolutional block attention mechanism to encode the grayscale fusion image in 4 layers to obtain the convolutional coding features;

[0077] Step Q2: Swin encoding, use the Swin-Unet encoder to encode the grayscale fusion image in 4 layers to obtain a variable voltage encoding feature with a dimension of (W / 32, H / 32, 16C);

[0078] Step Q3: Feature fusion decoding, after transposing the convolutional coding features, concatenate them with the transformer coding features, and perform 1*1 convolution and batch normalization to obtain the primary decoding result, whose dimension is (W / 16, H / 16, 4C);

[0079] Step Q4: Feature selection, use the channel attention SE module (Squeeze-and-Excitation module) to select features from the primary decoding results to obtain primary selected features;

[0080] Step Q5: Multi-layer decoding, perform 3-layer upsampling on the primary selected features to obtain the final decoded features with dimensions (W, H, C). Use the activation function to activate the final decoded features to obtain the segmented image.

[0081] By performing the above operations, the correlation information between the space and channels of the fused image is captured through convolutional attention encoding, and the global and local information of the fused image is captured through the Swin-Unet encoder. The two types of information are transposed and convolved and then connected and decoded to obtain a more accurate lesion location for precise edge positioning.

[0082] Embodiment 5: This embodiment is based on the above embodiment. In step S1, a brightness component image is obtained through an RGB image. Specifically, the following steps are included:

[0083] Step S11: Represent all pixel information of the RGB image in the form of three channels (R, G, B), where R, G, B∈[0,255];

[0084] Step S12: Convert the brightness of each pixel:

[0085] ;

[0086] In the formula, Represents the brightness value of the pixel;

[0087] Step S13: recombining all brightness values ​​into a new single-channel image to obtain a brightness component image.

[0088] Embodiment 6: This embodiment is based on the above embodiment. In step S31, the CAM channel attention module processes, specifically including the following steps:

[0089] Step S311: performing maximum pooling processing and multi-layer perceptron processing on the features and then performing channel expansion to obtain maximum pooling features;

[0090] Step S312: performing average pooling processing and multi-layer perceptron processing on the features and then performing channel expansion to obtain average pooling features, wherein the multi-layer perceptron in step S312 and the multi-layer perceptron in step S311 are two shared fully connected layers;

[0091] Step S313: The maximum pooling feature and the average pooling feature are added element by element and activated using the sigmoid function to obtain the primary attention feature.

[0092] Embodiment 7: This embodiment is based on the above embodiment, and the training data set is collected:

[0093] Collect the skin disease image dataset PH² as a training dataset;

[0094] Dataset Overview:

[0095] Image type: dermoscopic image;

[0096] Number of images: 200;

[0097] Resolution: 768×560 pixels;

[0098] Image format: JPEG;

[0099] Dataset composition:

[0100] Each image in the PH² dataset corresponds to a skin lesion area, including the following three types of skin lesions:

[0101] Common Nevus;

[0102] Atypical Nevus;

[0103] Melanoma

[0104] Each image is accompanied by relevant annotations and tags, including:

[0105] Manually segmented lesion area: used for image segmentation tasks;

[0106] Diagnostic information of skin lesions: the type of lesion (benign or malignant) is noted;

[0107] Feature information: Contains feature information about color, shape, texture, etc., which is used for segmentation tasks.

[0108] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0109] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

[0110] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. An intelligent nursing edge positioning and adjustment system based on image recognition, characterized in that: It includes an image acquisition module, an intelligent image fusion module, an intelligent image segmentation module, an edge positioning adjustment module and a user interface module; The image acquisition module uses a color camera to capture an RGB image of the patient's lesion, uses a depth camera to capture a depth image of the patient's lesion, and sends the RGB image and the depth image to the intelligent image fusion module; The intelligent image fusion module uses a multi-scale feature fusion method based on multi-dimensional dynamic convolution to fuse the RGB image and the depth image, obtains a grayscale fusion image and sends it to the intelligent image segmentation module; The multi-scale feature fusion method based on multi-dimensional dynamic convolution specifically includes the following steps: Step S1: Image conversion, converting the RGB image to the YCbCr space to obtain the brightness component image Y of the RGB image, and converting the depth image in a depth-to-brightness conversion manner to obtain the brightness image D of the depth image; Step S2: Multi-dimensional dynamic convolution, perform multi-dimensional attention dynamic convolution on the brightness component map Y and the brightness map D respectively, and obtain the brightness component convolution map and the brightness convolution map respectively: ； In the formula, represents the Gaussian error linear unit activation function, represents the normalization operation, It stands for multi-dimensional attention dynamic convolution, which includes four dimensions of attention dynamic convolution, namely spatial dimension, input channel dimension, output channel dimension and convolution kernel number dimension. The spatial dimension corresponds to the spatial position of each convolution kernel, the input channel dimension corresponds to the channel dimension of the input data, the output channel dimension corresponds to the channel dimension of the output data, and the convolution kernel number dimension corresponds to the weights between different convolution kernels. Represents the input image, namely the brightness component image Y and the brightness image D, Represents the output image, i.e., the brightness component convolution map and the brightness convolution map; Step S3: multimodal feature activation, performing multimodal feature activation on the brightness component convolution map and the brightness convolution map respectively, and obtaining brightness component activation features and brightness activation features respectively, specifically comprising the following steps: Step S31: The brightness component convolution map and the brightness convolution map are collectively referred to as the input image. The input image is subjected to LN normalization processing, and the window-based multi-head self-attention processing and channel attention module processing are respectively performed. The two processing results are added to the original input image to obtain the primary attention feature: ； In the formula, Represents the input image, namely the brightness component convolution map and the brightness convolution map, represents LN normalization, represents window-based multi-head self-attention processing, represents the channel attention module processing, represents primary attention features; Step S32: Perform LN normalization on the primary attention features and perform multi-layer perceptron processing, and add the processing results to the primary attention features to obtain preliminary activation features: ； In the formula, represents multi-layer perceptron processing, represents the initial activation feature; Step S33: Perform LN normalization on the preliminary activation features and perform overlapping attention module processing, and add the processing result to the preliminary activation features to obtain secondary attention features: ； In the formula, represents overlapping cross-attention module processing, represents secondary attention features; Step S34: perform LN normalization on the secondary attention feature and perform multi-layer perceptron processing, and add the processing result to the secondary attention feature to obtain the secondary activation feature: ； In the formula, represents multi-layer perceptron processing, represents the secondary activation features, namely, the brightness component activation features and the brightness activation features; Step S4: image fusion, fusing and activating the brightness component activation feature and the brightness activation feature to obtain a brightness fusion image; Step S5: grayscale image conversion, converting the brightness fusion image in a brightness-to-grayscale conversion manner to obtain a grayscale fusion image; The intelligent image segmentation module uses a dual encoder network based on multi-dimensional information extraction, captures the spatial and channel correlation information of the fused image through convolutional attention coding, captures the global and local information of the fused image through the Swin-Unet encoder, and performs convolution and feature selection on the two types of information and then performs concatenation decoding to obtain a segmented image, and finally sends the segmented image to the edge positioning adjustment module; The dual encoder network based on multidimensional information extraction specifically includes the following processing steps: Step Q1: CBAM encoding, use the encoder CBAM based on the convolutional block attention mechanism to encode the grayscale fusion image in n layers to obtain the convolutional coding features; Step Q2: Swin encoding, use the Swin-Unet encoder to encode the grayscale fusion image in n layers to obtain the variable voltage coding feature; Step Q3: Feature fusion decoding, after transposing the convolutional coding features, concatenate them with the transformer coding features, and perform 1*1 convolution and batch normalization to obtain the primary decoding result; Step Q4: Feature selection, use the channel attention SE module (Squeeze-and-Excitation module) to select features from the primary decoding results to obtain primary selected features; Step Q5: multi-layer decoding, perform n-1 layer upsampling on the primary selected features to obtain the final decoded features, and use the activation function to activate the final decoded features to obtain the segmented image; The edge positioning adjustment module determines the edge position and adjustment direction according to the segmented image to obtain the edge positioning adjustment result; The user interface module allows medical staff to view the results of the edge positioning adjustment, and the medical staff can accurately edge the patient's lesion based on the results. It also allows technical personnel to adjust parameters of all modules and view information.

Images