A method for locating the position of the renal pelvis in a renal pelvis ultrasound image

Abstract

The present invention discloses a method for locating the position of the renal pelvis in a renal pelvis ultrasound image, comprising the following steps: obtaining and processing an ultrasound image of the renal pelvis, constructing a basic data set and an enhanced data set, constructing a multi-scale deep learning model, inputting the basic data set into the multi-scale deep learning model for processing, obtaining channel attention features and spatial attention features, fusing the channel attention features and spatial attention features to obtain a fusion feature of the current layer, processing the enhanced data set and the fusion feature of the current layer to obtain the pixel coordinates of the renal pelvis in the ultrasound image, which are used to accurately locate the position of the renal pelvis in the ultrasound image of the renal pelvis. By constructing a multi-scale deep learning model, the present invention effectively extracts kidney features at different scales, overcomes the influence of ultrasound imaging angle and depth on the ultrasound image quality of the renal pelvis, and significantly improves the accuracy of renal pelvic puncture positioning.

Description

Technical Field

[0001] The present invention relates to the technical fields of ultrasonic image intelligent recognition and urological surgery, and in particular to a method for locating the position of a renal pelvis in a renal pelvis ultrasonic image. Background Art

[0002] Renal pelvis puncture is one of the commonly used diagnostic and treatment methods in urology, mainly used to treat diseases such as kidney stones and hydronephrosis. After the introduction of ultrasound equipment into clinical practice, ultrasound-guided renal pelvis puncture has become the preferred puncture method in clinical practice due to its real-time nature, accuracy, and safety. Through ultrasound guidance, doctors can directly visualize the dynamic images of the kidneys and surrounding tissues, accurately locate the puncture target, and reduce the occurrence of mispunctures and complications. However, ultrasound-guided renal pelvis puncture technology requires a high level of operating experience from the doctor and requires long-term professional training and practical accumulation. Therefore, there is an urgent need for an intelligent auxiliary technology to help doctors quickly and accurately locate the target during the puncture process, thereby improving the success rate and safety of the puncture.

[0003] The Chinese invention patent with publication number CN 202859244 U proposes a renal pelvis puncture needle and designs a model of the renal pelvis puncture needle. The main content is to design a renal pelvis puncture needle: the puncture needle includes an outer needle tube, an outer needle seat, an inner needle core and an inner needle seat, etc. The safety and convenience of puncture are improved by designing filter holes and grooves on the side of the inner needle core.

[0004] Chinese invention patent publication number CN 117197098 A proposes a deep learning-based automatic segmentation method for pediatric kidneys and pyelonephritis in computed tomography (CT) images. The method involves the following steps: S1, local feature extraction: local features of the input image are extracted using local feature extraction; S2, global feature extraction: the input from the local feature extraction stage is used as the input for the global feature extraction stage to extract global feature information; S3, decoding and upsampling: the output from the global feature extraction stage is used as input, and features are fused with the output from the local feature extraction stage. A designed adaptive attention module is then used to enable the model to learn important information from the fused features and filter out redundant information; S4, segmentation: convolution is used to convert the input into the desired category for segmentation.

[0005] The Chinese invention patent with publication number CN117557577A proposes a renal pelvis and ureter segmentation method and system based on plain scan CT and deep learning, and designs deep learning and renal pelvis segmentation. The method includes the following steps: S1, obtaining a plain scan CT image of the urinary system; S2, using the plain scan CT image as input, using the Swin-UNet target detection model based on the attention mechanism to process the plain scan CT image; S3, the target detection model includes several layers of neural networks, and each layer of the neural network performs downsampling, windowing and neighbor comparison operations in turn to obtain a feature map of each layer. ; S4. Starting from the smallest feature map, upsampling, windowing and neighbor comparison operations are performed in sequence, and then connected with the high-level feature map to determine whether each window block in the high-level feature map belongs to a certain organ; S5. Repeat multiple times to obtain the positions of the bilateral kidneys in the input image, and provide the relevant positions to the two segmentation models; S6. The first segmentation model is used to segment the renal pelvis and the upper part of the ureter to obtain the first segmentation result, and the second segmentation model is used to segment the middle and lower parts of the ureter to obtain the second segmentation result, and the first segmentation result and the second segmentation result are added together to obtain the renal pelvis structure and the ureter structure.

[0006] However, the above technology has the following limitations:

[0007] When acquiring ultrasound images of a patient's kidneys, electromagnetic interference and tissue scattering noise are present. Noisy ultrasound images can impair the physician's accurate judgment of the puncture target and increase the risk of puncture complications. The aforementioned technologies lack solutions specifically designed to improve ultrasound image quality, making it difficult to ensure accurate punctures.

[0008] Renal anatomy varies from patient to patient, and ultrasound imaging at varying angles and depths can lead to inconsistent image presentation. The aforementioned technologies fail to account for these variations and lack intelligent assisted positioning and navigation capabilities.

[0009] Existing renal pelvic puncture systems lack standardized and intelligent support systems. Training an AI model that can accurately identify a patient's renal pelvis location and plan the optimal puncture path requires a large amount of labeled data and clinical validation. This approach consumes significant human and material resources and poses challenges in data quality and standardization. Summary of the Invention

[0010] In view of the deficiencies of the prior art, the present invention provides a method for locating the position of the renal pelvis in a renal pelvis ultrasound image to solve the problems raised in the above background technology.

[0011] To achieve the above object, the present invention provides the following technical solution: a method for locating the position of the renal pelvis in a renal pelvis ultrasound image, comprising the following steps:

[0012] Step S1: obtaining an ultrasound image of the renal pelvis;

[0013] Step S2: constructing a deep denoising autoencoder; processing the ultrasound image of the renal pelvis in step S1 by the deep denoising autoencoder to obtain a low-noise ultrasound image of the renal pelvis;

[0014] Step S3: Processing the low-noise renal pelvis ultrasound image in step S2 to construct a basic dataset and an enhanced dataset;

[0015] Step S4: constructing a multi-scale deep learning model; inputting the basic data set into the multi-scale deep learning model for processing, wherein the basic data set is first subjected to feature branch separation processing in the multi-scale deep learning model to obtain features of low-noise renal pelvis image data, and then the features of the low-noise renal pelvis image data are processed to obtain channel attention features and spatial attention features, and the channel attention features and the spatial attention features are fused to obtain the fusion features of the current layer;

[0016] Step S5: Constructing a deep learning model based on self-supervised learning; the deep learning model based on self-supervised learning includes a multi-scale feature extraction module and a renal pelvis localization module. The multi-scale feature extraction module processes the enhanced dataset to obtain multi-scale features; the multi-scale features and the fusion features of the current layer obtained in step S4 are input into the renal pelvis localization module for processing to obtain the pixel coordinates of the renal pelvis in the ultrasound image;

[0017] Step S6: Analyze the ultrasound image of the renal pelvis based on the pixel coordinates of the renal pelvis in the ultrasound image, and locate the position of the renal pelvis in the ultrasound image of the renal pelvis.

[0018] Furthermore, the ultrasound image of the renal pelvis in step S1 is specifically processed as follows:

[0019] Place the ultrasound probe on the patient's abdomen, scan horizontally along the patient's abdominal intercostal space or abdominal midline area, gradually move the ultrasound probe to the renal pelvis area, and use the ultrasound probe to detect different angles and depths of the renal pelvis area to obtain an ultrasound image of the renal pelvis.

[0020] Furthermore, in step S2, the low-noise renal pelvis ultrasound image is obtained by:

[0021] The ultrasound image of the renal pelvis is input into the encoder of the deep denoising autoencoder, and the deep feature map is obtained after the encoder's data preprocessing and the first and second convolutional layers. , expressed as:

[0022] ;

[0023] Where, represents the latent space encoding operation; represents the weight matrix in the encoder; represents the bias of the encoder average pooling layer; Ultrasound image showing the renal pelvis;

[0024] The depth feature map The input is sent to the decoder of the deep denoising autoencoder, and a low-noise renal pelvis ultrasound image is generated after upsampling and the first and second deconvolution layers. , expressed as:

[0025] ;

[0026] Where, Represents an upsampling operation; represents the weight matrix in the decoder; Represents the bias of the decoder average pooling layer.

[0027] Furthermore, in step S3, the basic dataset and enhanced dataset are constructed. The specific process is as follows:

[0028] Defining low-noise renal pelvis ultrasound images The total number of samples generated is , define the index sequence ; represents the total number of low-noise renal pelvis ultrasound image samples; Represents an index sequence set of low-noise renal pelvis ultrasound image samples; Randomly select an index sequence set of low-noise renal pelvis ultrasound image samples Evenly distribute 50% of the elements in the set and record it as , the remaining elements form the set ;

[0029] Based on the collection , select the corresponding low-noise renal pelvis ultrasound image samples from the index sequence set of low-noise renal pelvis ultrasound image samples to construct a basic data set ;

[0030] Based on the collection , select the corresponding low-noise renal pelvis ultrasound image samples from the index sequence set of low-noise renal pelvis ultrasound image samples to construct an enhanced data set .

[0031] Furthermore, the fusion feature of the current layer in step S4 is specifically processed as follows:

[0032] The basic dataset After performing image preprocessing on the low-noise renal pelvis ultrasound image samples, a low-noise renal pelvis ultrasound image after image preprocessing is obtained, and then the low-noise renal pelvis ultrasound image after image preprocessing is input into the feature branch separation to obtain the features of the low-noise renal pelvis image data;

[0033] The features of low-noise renal pelvis image data are expressed as ;Will Divided into high-resolution feature maps , medium-resolution feature map and low-resolution feature maps Three characteristic paths, expressed as:

[0034] ;

[0035] Where, represents the feature branch separation function;

[0036] The specific steps of feature branch separation function expansion are expressed as:

[0037] ;

[0038] ;

[0039] ;

[0040] ;

[0041] Where, Represents the characteristics of the input low-noise renal pelvis image data of The preprocessed feature map obtained after the convolution operation; express Convolution operation; express Convolution operation; express Convolution operation; express Convolution operation;

[0042] The high-resolution feature map , medium-resolution feature map and low-resolution feature maps The input is processed in the variable convolution module. The specific process is as follows:

[0043] The variable convolution module consists of a first variable convolution module, a second variable convolution module and a third variable convolution module;

[0044] The high-resolution feature map , medium-resolution feature map and low-resolution feature maps The outputs are input into the first variable convolution module, the second variable convolution module and the third variable convolution module for processing respectively, and the outputs of the first variable convolution module, the second variable convolution module and the third variable convolution module are obtained respectively, which are expressed as:

[0045] ;

[0046] ;

[0047] ;

[0048] Where, 、 and Represents high-resolution feature maps In position The value at, medium resolution feature map In position The value at and the low-resolution feature map In position The value at ; i and j represent the horizontal and vertical coordinates of the resolution feature map respectively; Represents the negative value of the convolution kernel radius; Represents the position in the convolution kernel The weight value of and Represent the vertical and horizontal positions in the convolution kernel respectively; and Indicates the location The learned spatial offset; represents the modulation factor; represents the convolution kernel radius;

[0049] The output of the first variable convolution module, the output of the second variable convolution module and the output of the third variable convolution module are fused through feature splicing to obtain the output of feature splicing, and the output of feature splicing is input into the channel attention module and the spatial attention module for processing to obtain the channel attention module. Layer Features and spatial attention Layer Features ;

[0050] The channel attention module is used to enhance the output of feature splicing and obtain the channel attention Layer Features , expressed as:

[0051] ;

[0052] Where, represents the activation function; represents a multilayer perceptron; and Represents average pooling and maximum pooling operations respectively; Represents the output of feature concatenation; for 、 and The splicing features, ; Represents feature splicing;

[0053] The spatial attention module is used to enhance the output of feature splicing to obtain the spatial attention Layer Features , expressed as:

[0054] ; Use feature pyramid network to focus on the channel Layer Features and spatial attention Layer Features Perform fusion to obtain the fusion features of the current layer, which is expressed as:

[0055] ;

[0056] Where, Represents the fusion features of the current layer; Conv( is the convolution operation; ; Represents the fusion features of the current first layer; Indicates the current Layer fusion features; Indicates the current Layer fusion features; Indicates the +1 layer of fused features.

[0057] Furthermore, the pixel coordinates of the renal pelvis in the ultrasound image in step S5 are specifically obtained as follows:

[0058] The dataset will be enhanced Input the deep learning model based on self-supervised learning, which includes a multi-scale feature extraction module and a renal pelvis positioning module; the enhanced dataset is trained by the multi-scale feature extraction module , and then repeat the feature branch separation step in step S4 to obtain multi-scale features ;

[0059] The renal pelvis localization module consists of two fully connected layers, which integrate the multi-scale features and the fusion features of the current layer Input the renal pelvis positioning module for processing and output the position of the renal pelvis pixel point , which is the pixel coordinate of the renal pelvis in the ultrasound image.

[0060] Compared with the existing technology, the present invention has the following beneficial effects:

[0061] (1) The present invention significantly improves the quality of renal ultrasound images through a deep denoising autoencoder, removes electromagnetic interference and tissue scattering noise, and thus improves the accuracy and safety of puncture path positioning.

[0062] (2) By constructing a multi-scale deep learning model, the present invention can effectively extract kidney features at different scales, overcome the influence of different ultrasound imaging angles and depths on the ultrasound image quality of the renal pelvis, and significantly improve the accuracy of renal pelvic puncture positioning.

[0063] (3) The present invention adopts a self-supervised learning method to achieve high-precision renal pelvis positioning and puncture path planning based on a smaller data set, significantly reducing the cost of data collection and deep denoising autoencoder training. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] Figure 1 Flowchart of the present invention. DETAILED DESCRIPTION

[0065] like Figure 1 As shown, the present invention provides a technical solution: a method for locating the position of the renal pelvis in a renal pelvis ultrasound image, comprising the following steps:

[0066] Step S1: obtaining an ultrasound image of the renal pelvis;

[0067] Step S2: constructing a deep denoising autoencoder; processing the ultrasound image of the renal pelvis in step S1 by the deep denoising autoencoder to obtain a low-noise ultrasound image of the renal pelvis;

[0068] Step S3: Processing the low-noise renal pelvis ultrasound image in step S2 to construct a basic dataset and an enhanced dataset;

[0069] Step S4: constructing a multi-scale deep learning model; inputting the basic data set into the multi-scale deep learning model for processing, wherein the basic data set is first subjected to feature branch separation processing in the multi-scale deep learning model to obtain features of low-noise renal pelvis image data, and then the features of the low-noise renal pelvis image data are processed to obtain channel attention features and spatial attention features, and the channel attention features and the spatial attention features are fused to obtain the fusion features of the current layer;

[0070] Step S5: Constructing a deep learning model based on self-supervised learning; the deep learning model based on self-supervised learning includes a multi-scale feature extraction module and a renal pelvis localization module. The multi-scale feature extraction module processes the enhanced dataset to obtain multi-scale features; the multi-scale features and the fusion features of the current layer obtained in step S4 are input into the renal pelvis localization module for processing to obtain the pixel coordinates of the renal pelvis in the ultrasound image;

[0071] Step S6: Analyze the ultrasound image of the renal pelvis based on the pixel coordinates of the renal pelvis in the ultrasound image, and locate the position of the renal pelvis in the ultrasound image of the renal pelvis.

[0072] The specific process of the ultrasound image of the renal pelvis in step S1 is as follows:

[0073] Place the ultrasound probe on the patient's abdomen, scan horizontally along the intercostal space or the midline of the abdomen, and gradually move the ultrasound probe to the renal pelvis area. Use the ultrasound probe to detect different angles and depths of the renal pelvis area to obtain an ultrasound image of the renal pelvis.

[0074] The specific process of obtaining the low-noise renal pelvis ultrasound image in step S2 is as follows:

[0075] Ultrasound images of the renal pelvis are susceptible to interference from factors such as instrument accuracy, patient physiological activity, and environmental factors, which can generate noise. This can blur key details such as the kidney's internal microstructure and vascular texture, hindering doctors' ability to accurately assess the severity of the disease. To address this, a deep denoising autoencoder was constructed to accurately remove noise from renal pelvis ultrasound images, providing a clear image basis for subsequent diagnosis.

[0076] The encoder of the deep denoising autoencoder consists of the first convolutional layer and the second convolutional layer;

[0077] The decoder of the deep denoising autoencoder consists of an upsampling layer, a first deconvolution layer, and a second deconvolution layer;

[0078] The ultrasound image of the renal pelvis is input into the encoder, and the depth feature map is obtained after the encoder's data preprocessing and the first convolution layer and the first convolution layer operation. , expressed as:

[0079] ;

[0080] Where, represents the latent space encoding operation; represents the weight matrix in the encoder; represents the bias of the encoder average pooling layer; Ultrasound image showing the renal pelvis;

[0081] The depth feature map In the input decoder, a low-noise renal pelvis ultrasound image is generated after upsampling and the first and second deconvolution layers. , expressed as:

[0082] ;

[0083] Where, Represents an upsampling operation; represents the weight matrix in the decoder; Represents the bias of the decoder average pooling layer.

[0084] Among them, in step S3, the basic dataset and enhanced dataset are constructed, and the specific process is as follows:

[0085] The total number of samples generated by low-noise renal pelvis ultrasound images is defined as , define the index sequence ; represents the total number of samples of low-noise renal pelvis ultrasound images; Represents the index sequence set of low-noise renal pelvis ultrasound image samples; the index sequence set of random low-noise renal pelvis ultrasound image samples Evenly distribute 50% of the elements in the set and record it as , the remaining elements form the set ;

[0086] Based on the collection , select the corresponding low-noise renal pelvis ultrasound image samples from the index sequence set of low-noise renal pelvis ultrasound image samples to construct a basic data set ;

[0087] Based on the collection , select the corresponding low-noise renal pelvis ultrasound image samples from the index sequence set of low-noise renal pelvis ultrasound image samples to construct an enhanced data set ;

[0088] In the basic data set In the enhanced dataset, all low-noise renal pelvis ultrasound image samples contain clear manual annotation information to ensure that the deep learning model can learn specific pathological features during the initial training process; In the data enhancement process, images of different angles, depths and pathological conditions are processed to enhance the data set. It does not include manual annotation and is used for semi-supervised learning tasks to increase the generalization ability of multi-scale deep learning models; data enhancement techniques include operations such as rotation, translation, mirror reversal, and contrast adjustment to simulate the changes of renal ultrasound images obtained in real scenes under various conditions.

[0089] Among them, the fusion feature of the current layer in step S4 is as follows: the basic data set After performing image preprocessing on the low-noise renal pelvis ultrasound image samples, a low-noise renal pelvis ultrasound image after image preprocessing is obtained, and then the low-noise renal pelvis ultrasound image after image preprocessing is input into the feature branch separation to obtain the features of the low-noise renal pelvis image data;

[0090] The features of low-noise renal pelvis image data are expressed as ;Will Divided into high-resolution feature maps , medium-resolution feature map and low-resolution feature maps The three feature paths capture the detail information, structural information, and global semantic information of low-noise renal pelvis ultrasound images, respectively, and are expressed as:

[0091] ;

[0092] Where, Represents the feature branch separation function; it is the core operation of feature branch separation;

[0093] The specific steps of feature branch separation function expansion are expressed as:

[0094] ;

[0095] ;

[0096] ;

[0097] ;

[0098] Where, Represents the characteristics of the input low-noise renal pelvis image data of The preprocessed feature map obtained after the convolution operation; express Convolution operation; express Convolution operation; express Convolution operation; express Convolution operation;

[0099] The high-resolution feature map , medium-resolution feature map and low-resolution feature maps The input is processed in the variable convolution module. The specific process is as follows:

[0100] The variable convolution module consists of a first variable convolution module, a second variable convolution module and a third variable convolution module;

[0101] The high-resolution feature map , medium-resolution feature map and low-resolution feature maps The outputs are input into the first variable convolution module, the second variable convolution module and the third variable convolution module for processing respectively, and the outputs of the first variable convolution module, the second variable convolution module and the third variable convolution module are obtained respectively, which are expressed as:

[0102] ;

[0103] ;

[0104] ;

[0105] Where, 、 and Represents high-resolution feature maps In position The value at, medium resolution feature map In position The value at and the low-resolution feature map In position The value at ; i and j represent the horizontal and vertical coordinates of the resolution feature map respectively; Represents the negative value of the convolution kernel radius; Represents the position in the convolution kernel The weight value of and Represent the vertical and horizontal positions in the convolution kernel respectively; and Indicates the location The learned spatial offset; represents the modulation factor; represents the convolution kernel radius;

[0106] The output of the first variable convolution module, the output of the second variable convolution module, and the output of the third variable convolution module are fused through feature splicing to obtain the output of feature splicing. The output of feature splicing is input into the channel attention module and the spatial attention module for processing to obtain the channel attention feature. and spatial attention features ;

[0107] The channel attention module is used to enhance the output of feature splicing and obtain the channel attention Layer Features , expressed as:

[0108] ;

[0109] Where, represents the activation function; represents a multilayer perceptron; and Represents average pooling and maximum pooling operations respectively; Represents the input feature map; for 、 and The splicing features, ; Represents feature splicing;

[0110] Due to the uneven spatial distribution of low-noise renal pelvis ultrasound image quality, the clarity and reliability of different areas of the low-noise renal pelvis ultrasound image are different, and it is necessary to adaptively adjust the attention. Therefore, the present invention uses a spatial attention module to enhance the output of feature splicing to obtain the spatial attention first. Layer Features , expressed as:

[0111] ;

[0112] Use Feature Pyramid Network (FPN) to focus on the channel Layer Features and spatial attention Layer Features Perform fusion to obtain the fusion features of the current layer, which is expressed as:

[0113] ;

[0114] Where, Represents the fusion features of the current layer; Conv( It is a convolution operation used to smooth the fused features; ; Represents the fusion features of the current first layer; Indicates the current Layer fusion features; Indicates the current Layer fusion features; Indicates the +1 layer of fusion features;

[0115] In the low-noise renal pelvis ultrasound image feature extraction task, since it is necessary to simultaneously ensure accurate segmentation of the puncture area, precise estimation of the path depth, clear positioning of tissue boundaries, and highlighting of key anatomical structures, a single loss function is difficult to fully guide the multi-scale deep learning model to learn these complex feature expressions. Therefore, a multi-component loss function is designed, which includes a segmentation loss function, a depth estimation loss function, a contour perception loss function, and an attention-guided loss function. Through the synergistic effect and dynamic weight adjustment of each loss term, the multi-scale deep learning model is guided to focus on multiple clinical key targets simultaneously during training, thereby obtaining a more comprehensive and robust feature representation and improving the accuracy and safety of puncture surgery. The fused features of the current layer are input into the segmentation loss, depth estimation loss, contour perception loss, and attention-guided loss for processing, respectively, to obtain the outputs of the segmentation loss, depth estimation loss, contour perception loss, and attention-guided loss, respectively. The outputs of the segmentation loss, depth estimation loss, contour perception loss, and attention-guided loss are added together to obtain the total loss function, which is expressed as:

[0116] ;

[0117] Where, represents the total loss function; represents the output of segmentation loss; represents the output of the depth estimation loss; represents the output of contour-aware loss; represents the output of the attention-guided loss; 、 、 、 They represent the output weight parameters of segmentation loss, depth estimation loss, contour perception loss, and attention-guided loss, respectively.

[0118] The pixel coordinates of the renal pelvis in the ultrasound image in step S5 are specifically obtained as follows:

[0119] The dataset will be enhanced Input into the multi-scale feature extraction module, and train the enhanced dataset through the multi-scale feature extraction module , and then repeat the feature branch separation step in step S4 to obtain multi-scale features , to provide support for subsequent positioning tasks; expressed as:

[0120] ;

[0121] Where, Representing multi-scale features In position The element value of Indicates that the input renal pelvis ultrasound image is at position Pixel value of

[0122] The renal pelvis localization module consists of two fully connected layers, which integrate the multi-scale features and the fusion features of the current layer Input the renal pelvis positioning module for processing and output the position of the renal pelvis pixel point , which is the pixel coordinate of the renal pelvis in the ultrasound image.

[0123] The deep learning model based on self-supervised learning includes image reconstruction tasks, contrastive learning tasks, and context prediction tasks. The image reconstruction task is completed through the autoencoder structure, and the input low-noise renal pelvis ultrasound image is reconstructed through the image reconstruction task to obtain the reconstructed input low-noise renal pelvis ultrasound image; the contrastive learning task is to train the deep learning model based on self-supervised learning to maintain feature consistency by inputting low-noise renal pelvis ultrasound images enhanced by different views, and obtain the trained self-supervised learning deep learning model; the context prediction task is to capture the structural features of the low-noise renal pelvis ultrasound image by predicting the relative position of the local area of ​​the reconstructed input low-noise renal pelvis ultrasound image;

[0124] In the fine-tuning stage, based on the structural features of low-noise renal pelvis ultrasound images, the enhanced dataset 10% of the data is randomly selected and input into the trained self-supervised learning deep learning model for self-supervised learning to achieve high-precision renal pelvis positioning; it is expressed as:

[0125] ;

[0126] Where, represents the self-supervised localization loss function; Indicates the total number of positions; Represents the weights of different pixel positions; Represents the predicted renal pelvis pixel set The position of the pixel; Indicates the first pixel in the real renal pelvis pixel set The position of the pixel;

[0127] A self-supervised localization loss function is used to train the deep learning model based on self-supervised learning. The self-supervised localization loss function includes the image reconstruction task loss function, the contrastive learning task loss function, and the context prediction task loss function. The self-supervised localization loss function can effectively learn the characteristic patterns in low-noise renal pelvis ultrasound images and provide high-quality feature representation for subsequent localization tasks. The loss calculation formula is expressed as:

[0128] ;

[0129] Where, represents the total loss function; Represents the loss function of the image reconstruction task; represents the loss function of contrastive learning task; Represents the context prediction task loss function.

[0130] While 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 these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A method for locating the position of the renal pelvis in a renal pelvis ultrasound image, characterized in that: The following steps are involved: Step S1: obtaining an ultrasound image of the renal pelvis; Step S2: constructing a deep denoising autoencoder; processing the ultrasound image of the renal pelvis in step S1 by the deep denoising autoencoder to obtain a low-noise ultrasound image of the renal pelvis; Step S3: Processing the low-noise renal pelvis ultrasound image in step S2 to construct a basic dataset and an enhanced dataset; Step S4: construct a multi-scale deep learning model; The basic data set is input into the multi-scale deep learning model for processing. The basic data set is first subjected to feature branch separation processing in the multi-scale deep learning model to obtain the features of the low-noise renal pelvis image data. The features of the low-noise renal pelvis image data are then processed to obtain channel attention features and spatial attention features. The channel attention features and spatial attention features are fused to obtain the fusion features of the current layer; Step S5: Constructing a deep learning model based on self-supervised learning; The deep learning model based on self-supervised learning includes a multi-scale feature extraction module and a renal pelvis localization module. The multi-scale feature extraction module processes the enhanced data set to obtain multi-scale features. The multi-scale features and the fusion features of the current layer obtained in step S4 are input into the renal pelvis localization module for processing to obtain the pixel coordinates of the renal pelvis in the ultrasound image. Step S6: Analyze the ultrasound image of the renal pelvis based on the pixel coordinates of the renal pelvis in the ultrasound image, and locate the position of the renal pelvis in the ultrasound image of the renal pelvis.

2. The method for locating the renal pelvis in a renal pelvis ultrasound image according to claim 1, characterized in that: The specific process of the ultrasound image of the renal pelvis in step S1 is as follows: Place the ultrasound probe on the patient's abdomen, scan horizontally along the patient's abdominal intercostal space or abdominal midline area, gradually move the ultrasound probe to the renal pelvis area, and use the ultrasound probe to detect different angles and depths of the renal pelvis area to obtain an ultrasound image of the renal pelvis.

3. The method for locating the renal pelvis in a renal pelvis ultrasound image according to claim 2, characterized in that: The specific process of obtaining the low-noise renal pelvis ultrasound image in step S2 is as follows: The ultrasound image of the renal pelvis is input into the encoder of the deep denoising autoencoder, and the deep feature map is obtained after the encoder's data preprocessing and the first and second convolutional layers. , expressed as: ； Where, represents the latent space encoding operation; represents the weight matrix in the encoder; represents the bias of the encoder average pooling layer; Ultrasound image showing the renal pelvis; The depth feature map The input is sent to the decoder of the deep denoising autoencoder, and a low-noise renal pelvis ultrasound image is generated after upsampling and the first and second deconvolution layers. , expressed as: ； Where, Represents an upsampling operation; represents the weight matrix in the decoder; Represents the bias of the decoder average pooling layer.

4. The method for locating the renal pelvis in a renal pelvis ultrasound image according to claim 3, wherein: In step S3, the basic dataset and enhanced dataset are constructed. The specific process is as follows: Defining low-noise renal pelvis ultrasound images The total number of samples generated is , define the index sequence ; represents the total number of low-noise renal pelvis ultrasound image samples; Represents an index sequence set of low-noise renal pelvis ultrasound image samples; Randomly select an index sequence set of low-noise renal pelvis ultrasound image samples Evenly distribute 50% of the elements in the set and record it as , the remaining elements form the set ; Based on the collection , select the corresponding low-noise renal pelvis ultrasound image samples from the index sequence set of low-noise renal pelvis ultrasound image samples to construct a basic data set ; Based on the collection , select the corresponding low-noise renal pelvis ultrasound image samples from the index sequence set of low-noise renal pelvis ultrasound image samples to construct an enhanced data set .

5. The method for locating the renal pelvis in a renal pelvis ultrasound image according to claim 4, characterized in that: The fusion feature of the current layer in step S4 is as follows: The basic dataset After performing image preprocessing on the low-noise renal pelvis ultrasound image samples, a low-noise renal pelvis ultrasound image after image preprocessing is obtained, and then the low-noise renal pelvis ultrasound image after image preprocessing is input into the feature branch separation to obtain the features of the low-noise renal pelvis image data; The features of low-noise renal pelvis image data are expressed as ;Will Divided into high-resolution feature maps , medium-resolution feature map and low-resolution feature maps Three characteristic paths, expressed as: ； Where, represents the feature branch separation function; The specific steps of feature branch separation function expansion are expressed as: ； ； ； ； Where, Represents the characteristics of the input low-noise renal pelvis image data of The preprocessed feature map obtained after the convolution operation; express Convolution operation; express Convolution operation; express Convolution operation; express Convolution operation; The high-resolution feature map , medium-resolution feature map and low-resolution feature maps The input is processed in the variable convolution module. The specific process is as follows: The variable convolution module consists of a first variable convolution module, a second variable convolution module and a third variable convolution module; The high-resolution feature map , medium-resolution feature map and low-resolution feature maps The first variable convolution module, the second variable convolution module, and the third variable convolution module are input for processing respectively, and the output of the first variable convolution module, the output of the second variable convolution module, and the output of the third variable convolution module are obtained respectively, which are expressed as: ； ； ； Where, 、 and Represent high-resolution feature maps In position The value at, medium resolution feature map In position The value at and the low-resolution feature map In position The value at ; i and j represent the horizontal and vertical coordinates of the resolution feature map respectively; Represents the negative value of the convolution kernel radius; Represents the position in the convolution kernel The weight value of and Represent the vertical and horizontal positions in the convolution kernel respectively; and Indicates the location The learned spatial offset; represents the modulation factor; represents the convolution kernel radius; The output of the first variable convolution module, the output of the second variable convolution module and the output of the third variable convolution module are fused through feature splicing to obtain the output of feature splicing, and the output of feature splicing is input into the channel attention module and the spatial attention module for processing to obtain the channel attention module. Layer Features and spatial attention Layer Features ; The channel attention module is used to enhance the output of feature splicing and obtain the channel attention Layer Features , expressed as: ； Where, represents the activation function; represents a multilayer perceptron; and Represents average pooling and maximum pooling operations respectively; Represents the output of feature concatenation; for 、 and The splicing features, ; Represents feature splicing; The spatial attention module is used to enhance the output of feature splicing to obtain the spatial attention Layer Features , expressed as: ； Using feature pyramid network to focus on channel Layer Features and spatial attention Layer Features Perform fusion to obtain the fusion features of the current layer, which is expressed as: ； Where, Represents the fusion features of the current layer; Conv( is the convolution operation; ; Represents the fusion features of the current first layer; Indicates the current Layer fusion features; Indicates the current Layer fusion features; Indicates the +1 layer of fused features.

6. The method for locating the renal pelvis in a renal pelvis ultrasound image according to claim 5, characterized in that: The pixel coordinates of the renal pelvis in the ultrasound image in step S5 are obtained by: The dataset will be enhanced Input the deep learning model based on self-supervised learning, which includes a multi-scale feature extraction module and a renal pelvis positioning module; the enhanced dataset is trained by the multi-scale feature extraction module , and then repeat the feature branch separation step in step S4 to obtain multi-scale features ; The renal pelvis localization module consists of two fully connected layers, which integrate the multi-scale features and the fusion features of the current layer Input the renal pelvis positioning module for processing and output the position of the renal pelvis pixel point , which is the pixel coordinate of the renal pelvis in the ultrasound image.

Images