An automated diagnostic method and system for excretory cystourethrography using deep learning

Abstract

This invention relates to the field of medical testing technology, and more particularly to an automated diagnostic method and system for voiding cystourethrography (VCUG) using deep learning. This method automatically locates and diagnoses the bladder, ureters, and urethra based on VCUG images. It not only enables automatic VUR grading but also diagnoses abnormalities in the bladder and urethra, thus broadening its clinical application. Furthermore, utilizing deep learning to build a model for automated diagnosis effectively reduces the influence of human factors on the results, improving accuracy. It also significantly assists physicians, especially junior physicians, in making timely and accurate diagnoses, helping to address issues such as low result repeatability between physicians of different experience levels and between different hospitals.

Description

Technical Field

[0001] This invention relates to the field of medical testing technology, and in particular to an automated diagnostic method and system for excretory cystourethrography using deep learning. Background Technology

[0002] Voiding cystourethrogram (VCUG) is one of the most common pediatric examinations and is considered the gold standard for depicting the anatomy and function of the lower urinary tract. Therefore, VCUG is suitable for investigating patients with various clinical conditions, including urinary tract infections, prenatal hydronephrosis, posterior urethral valves, and cloacal abnormalities. During filling, the shape, size, filling defects, and vesicoureteral reflux (VUR) of the bladder are assessed. During voiding, in addition to assessing reflux, any anatomical abnormalities or voiding dysfunctions of the bladder and urethra can be evaluated simultaneously. VCUG is considered the gold standard for various urological diagnoses, including VUR. Currently, although several guidelines have standardized the procedure of VCUG, the reported outcomes of VCUG cases are inconsistent.

[0003] In recent years, deep neural network models have been increasingly widely used in the field of medical imaging. These models can not only process large amounts of medical images efficiently, quickly, and automatically, reducing the workload of doctors, but also make image-based diagnosis, assessment, and image-guided intervention or treatment more objective, accurate, and intelligent. In the literature “Eroglu Y, Yildirim K, Çinar A, Yildirim M. Diagnosis and grading of vesicoureteral reflux on voiding cystourethrography images in children using a deep hybrid model. Computermethods and programs in biomedicine 2021; 210: 106369.”, Yesim et al. and in the literature “Li Z, Tan Z, Wang Z, et al. Development and multi-institutional validation of deep learning model for grading of vesicoureteral reflux on voiding cystourethrography: a retrospective multicenter study. EClinicalMedicine2024; 69: 102466.”, Li Zhanchi et al. proposed deep learning methods based on VCUG images for automatic VUR grading. However, these current methods can only perform VUR grading on a single VCUG image. Because VCUG examination is a dynamic process, typically involving multiple images, and requires attention to other abnormalities in the bladder and urethra in addition to grading VUR, existing deep learning methods still have limited coverage in practical clinical applications.

[0004] The applicant disclosed a method for automatic grading of vesicoureteral reflux using deep learning in Chinese invention patent application (Publication No.: CN115762753A, Publication Date: 2023-03-07). The method includes the following steps: collecting a set of vesicoureteral reflux (VCUG) images from VUR patients; manually labeling the reflux areas to complete the dataset collection; using an ensemble deep learning method with voting to select the optimal result as the overall prediction label; training the model with the preprocessed training dataset to form the final VCUG automatic grading model; inputting new images for prediction, performing the same image data preprocessing to obtain a standard vesicoureteral reflux grading map; and automatically grading the obtained standard vesicoureteral reflux grading map according to the VCUG automatic grading model. This method does not involve processing multiple images. Summary of the Invention

[0005] To address the shortcomings of existing technologies and expand the clinical applicability of intelligent VCUG grading, this invention provides a deep learning-based automated diagnostic method for voiding cystourethrography. This method automatically locates and diagnoses the bladder, ureter, and urethra based on VCUG images, enabling efficient and accurate automated diagnosis of voiding cystourethrography.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] An automated diagnostic method for excretory cystourethrography using deep learning, the method comprising the following steps:

[0008] Step 1: Data Acquisition and Image Selection

[0009] Images of children undergoing excretory cystourethrography were collected, and images with poor quality were excluded;

[0010] Step 2, Image Annotation and Dataset Construction:

[0011] Based on the selected VCUG photos, the bladder, urethra, and areas with vesicoureteral reflux are segmented and classified. If there is a discrepancy between the two classifications, a senior physician will make a final classification to complete the collection of the dataset.

[0012] Step 3: Deep learning model construction:

[0013] An automatic diagnostic model for VCUG is constructed using deep learning methods. The model mainly consists of two parts:

[0014] 1) Segmentation and classification based on a single VCUG image: A unified transformer-based architecture is used to handle classification and segmentation tasks simultaneously; the input is a single VCUG image, and the output is a lesion mask and diagnosis;

[0015] 2) Construct an automatic segmentation and diagnosis model for VCUG based on multiple VCUG images of a single patient: Extract high-dimensional feature representations of each VCUG image based on the output of the first part of the model, and superimpose these disease-related representations into a matrix; then design a multi-task attention layer to dynamically learn the contribution of each high-dimensional representation, and finally predict whether the patient has bladder or urethral abnormalities and determine the grade of ureteral reflux.

[0016] Step 4: Testing the deep learning model:

[0017] Multiple VCUG images from a new single patient are input, and image preprocessing is performed first. Then, the standardized VCUG images are input into the constructed automatic diagnostic model. The model automatically segments the bladder, urethra, and ureter regions and diagnoses lesions in the segmented regions.

[0018] Preferably, the image data collection includes: collecting a sufficient number of images of children undergoing VCUG examination; the labels for training the convolutional neural network model include: senior physicians segmenting the bladder, urethra, and ureter regions in a single VCUG image and making a diagnosis, the diagnosis results including whether the bladder and urethra are abnormal, and the degree of vesicoureteral reflux; for multiple VCUG images of a single person, a final diagnosis is made based on the results of each VCUG image, the diagnosis results including whether the bladder and urethra are abnormal, and the degree of vesicoureteral reflux.

[0019] Preferably, the automatic diagnostic model based on excretory cystourethrography (VCUG) processes the input VCUG images as follows: using the automatic diagnostic network model of excretory cystourethrography, each VCUG image is first segmented into the bladder, urethra, and ureter, and then the segmented regions are classified; next, for multiple VCUG images of a single person, the model integrates the segmentation and diagnostic results of each VCUG image to calculate the contribution, and finally makes a final diagnosis, which includes whether the bladder and urethra are abnormal, and the degree of vesicoureteral reflux.

[0020] Preferably, the preprocessing of each image data in the input VCUG includes: cropping and removing edge annotations, filling the edges of the image with black to make the image shape a square, then uniformly scaling the image to a size of 224×224 pixels, finally normalizing the pixel values ​​of the image to the range of 0~1, and standardizing the image according to the channel, i.e., affine transformation, perspective transformation, irregular deformation, and adding random noise; preferably, each image is randomly rotated by 0~15°.

[0021] Preferably, the automated diagnostic method model for excretory cystourethrography includes:

[0022] 1) A dataset that collects VCUG images with segmentation and diagnostic results as training samples;

[0023] 2) Data augmentation techniques are used on the collected training sample dataset to increase the number and variety of images by transforming them, thereby improving the generalization ability of the network model and its robustness to image segmentation.

[0024] 3) Deep learning feature extraction networks;

[0025] 4) A model for the automatic diagnosis of excretory cystourethrography.

[0026] As a preferred approach, the deep learning models used in excretory cystourethrography (VCUG) are all based on convolutional neural networks for feature extraction. Taking the processed VCUG image as input, the feature extraction is performed through convolutional layers, pooling layers, and activation function layers to finally obtain a feature map with complex semantic information. The automatic diagnostic model for excretory cystourethrography includes: unfolding the feature map of the VCUG image to obtain the corresponding feature vector of the image, performing feature fusion through a fully connected layer, then calculating the probability of the image being classified into each level through a softmax function, and finally comparing the magnitude of the probabilities to determine the level.

[0027] Furthermore, the present invention also discloses the application of the method in designing automated diagnostic software for emptying cystourethrography.

[0028] Furthermore, the present invention also discloses an automatic diagnostic system employing the method described above, the system comprising:

[0029] The image acquisition module is used to acquire images of children undergoing excretory cystourethrography and to filter the images, excluding those with poor image quality.

[0030] The image annotation module is used to segment and classify the bladder, urethra, and ureter regions of the selected VCUG images and generate an annotated dataset.

[0031] The deep learning model module is used to build an automatic diagnostic model based on deep learning. The model includes a segmentation and classification module based on a single VCUG image, and a comprehensive diagnostic module based on multiple VCUG images of a single person. The model dynamically learns the contribution of each image through a multi-task attention layer and outputs diagnostic results, including whether the bladder and urethra are abnormal and the degree of vesicoureteral reflux.

[0032] The data preprocessing module is used to preprocess the input VCUG image, including cropping, filling, scaling and normalization operations;

[0033] The automatic diagnosis module receives multiple VCUG images from patients, performs image segmentation and diagnosis, and generates segmentation and lesion diagnosis results for the bladder, urethra and ureter regions by analyzing the high-dimensional features of multiple images.

[0034] Preferably, the deep learning model module is initialized using a ViT-base model pre-trained on the ImageNet dataset, and the mask self-supervised learning technique is used to improve the system's feature extraction capability for medical images and accelerate the model training process.

[0035] Preferably, the system further includes a feature fusion module, which fuses the feature vectors of each image through a fully connected layer, calculates the probability of classifying the image into each level using a softmax function, and finally makes a diagnostic level determination based on the probability magnitude.

[0036] This invention, by employing the aforementioned technical solution, not only enables automatic VUR grading but also allows for the diagnosis of bladder and urethral abnormalities, thus broadening its clinical application. Furthermore, utilizing deep learning to build a model for automatic diagnosis effectively reduces the influence of human factors on the results, improving accuracy. It also significantly assists physicians' judgment, especially junior physicians, enabling them to make timely and accurate diagnoses, thus helping to address issues such as low repeatability of results between physicians of different experience levels and between different hospitals. Specifically, the technical effects are as follows:

[0037] 1. Automation and efficient diagnostics

[0038] This invention presents an automated diagnostic system for voiding cystourethrography (VCUG) based on deep learning. By utilizing a deep learning model and Transformer architecture, it achieves automatic segmentation and classification of multiple VCUG images. This enables the system to rapidly process large amounts of image data, effectively reducing the workload of physicians and improving diagnostic efficiency. Compared to traditional manual annotation and judgment, this system can process multiple images and automatically generate diagnostic results in a short time, significantly improving the efficiency of clinical diagnosis.

[0039] 2. Improved diagnostic accuracy and consistency

[0040] The deep learning model of this invention, through pre-training and multi-task attention mechanisms, can comprehensively analyze high-dimensional features in multiple VCUG images and dynamically learn the contribution of each image to the diagnosis. This not only improves the diagnostic accuracy of bladder and urethral abnormalities and vesicoureteral reflux (VUR) grades, but also effectively reduces diagnostic inconsistencies caused by human factors, especially in the operation of junior or less experienced doctors, ensuring the reliability of diagnostic results.

[0041] 3. Comprehensive multi-task diagnostics

[0042] This invention designs a multi-task model that can simultaneously segment and diagnose the bladder, urethra, and ureter, and further enhances the comprehensiveness of diagnosis by combining the correlation information between multiple tasks. This ability to simultaneously perform multi-task segmentation and classification enables the system not only to detect single diseases but also to assess abnormalities in multiple related organs, providing more comprehensive diagnostic information and broadening the system's application scope.

[0043] 4. Enhanced generalization ability and robustness of the model

[0044] By employing data augmentation techniques (such as affine transformation, perspective transformation, and adding random noise) in data processing, this invention effectively enhances the model's generalization ability and robustness. When processing real clinical data, the system can better adapt to differences in image quality, angle, and deformation, ensuring stable and high-quality diagnostic results across various scenarios.

[0045] 5. Significant clinical application value.

[0046] By applying deep learning to the automated diagnosis of emptying cystourethrography, this invention significantly reduces diagnostic discrepancies among physicians of varying experience, effectively addressing the problem of poor repeatability caused by the strong subjectivity of human judgment. Simultaneously, this system can significantly assist junior physicians in making accurate diagnoses, improving the level of medical services and enabling more hospitals and doctors to benefit, especially in resource-scarce regions.

[0047] 6. Reduce human intervention and lower the misdiagnosis rate.

[0048] This invention uses a deep learning model to automatically segment and diagnose the bladder, urethra, and ureters, effectively reducing the risk of misdiagnosis caused by human intervention. Simultaneously, the standardized diagnostic process improves diagnostic consistency, reduces the likelihood of misdiagnosis and missed diagnosis, and provides patients with more reliable diagnostic services. Attached Figure Description

[0049] Figure 1 This is an example of a diagnostic model for reflux in the bladder, urethra, and ureter based on VCUG images.

[0050] Figure 2 This is a schematic diagram illustrating a construction model using VCUG image-based automated diagnostics as an example. Detailed Implementation

[0051] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the protection scope of the present invention.

[0052] An automated diagnostic method for excretory cystourethrography includes the following steps:

[0053] Step 1: Collect an image set of children undergoing VCUG examination;

[0054] Step 2: For all the acquired VCUG images, two senior pediatric urologists segmented and diagnosed the bladder, urethra, and ureter regions of each VCUG image. They also made a final diagnosis based on multiple VCUG images from a single person. If there was a discrepancy between the two diagnoses, a senior physician would make a final classification, thus completing the collection of the dataset.

[0055] Step 3: Construct an excretory cystourethrography (VCUG) model using a Transformer network. The model architecture mainly consists of two parts: single-image segmentation and classification, and single-image result diagnosis for individual VCUG. The single-image VCUG model uses a unified Transformer-based architecture to handle both classification and segmentation tasks simultaneously. It takes an image as input and outputs segmented regions (bladder, urethra, ureter) and performs diagnoses, including bladder abnormalities, urethral abnormalities, and grading of vesicoureteral reflux. The model is initialized using ViT-based weights pre-trained on ImageNet with masked self-supervision, which stabilizes training, accelerates convergence, and enhances the representational capabilities of the image-level model.

[0056] Diagnosis based on multiple VCUG images from a single patient: We use the weights of an image-level model from the single image level as its initial weights. We freeze these weights to extract high-dimensional feature representations for each image and superimpose these disease-related representations into a matrix. A multi-task attention layer is then designed to dynamically learn the contribution of each high-dimensional representation, ultimately predicting whether the patient has bladder or urethral abnormalities and determining the grade of ureteral reflux.

[0057] Step 4: Input new single-person multiple VCUG images for prediction. First, perform the same image data preprocessing to obtain a standard image. Based on the constructed automatic diagnostic model based on excretory cystourethrography, first, segment and diagnose the bladder, urethra, and ureter regions of a single VCUG image based on the single-image model. Then, integrate the results of each VCUG image and perform the final diagnosis of multiple VCUG images of a single person.

[0058] Among them, image data collection involves collecting a certain number of patient images that have undergone VCUG examination, and senior physicians will segment and diagnose the bladder, ureter and urethra regions.

[0059] An automated diagnostic model for excretory cystourethrography (VCUG) constructed using a deep learning model is named VCUG-DAM. Using the collected samples as a training set, this model is responsible for segmenting and diagnosing the lesion regions (bladder, urethra, and ureter) in a single VCUG image; and for making a final diagnosis based on the results of integrating multiple VCUG images from a single individual.

[0060] Furthermore, the image data collection of the present invention includes:

[0061] Collect a sufficient number of VCUG images from patients, excluding images of poor quality (e.g., blurry, with large overlapping annotations, or lacking kidney and ureteral regions). Senior physicians segment and diagnose the bladder, ureter, and urethral regions on the collected VCUG images;

[0062] The image datasets collected above will be used as training samples for building an automated diagnostic model based on excretory cystourethrography.

[0063] Furthermore, the processing of multiple VCUG images of a single person during the model construction process includes:

[0064] An automated diagnostic model based on excretory cystourethrography is employed: First, the image-based model uses a unified transformer-based architecture to simultaneously handle classification and segmentation tasks. It takes an image as input and outputs segmented regions (bladder, urethra, ureter) and provides diagnoses, including bladder abnormalities, urethral abnormalities, and the classification of vesicoureteral reflux. The model is initialized using masked self-supervised ViT-based weights pre-trained on ImageNet, stabilizing training, accelerating convergence, and enhancing the representational capabilities of the image-level model.

[0065] Subsequently, the patient-based model uses weights from the image-level model as its initial weights. We freeze these weights to extract high-dimensional feature representations for each image and stack these disease-related representations into a matrix. A multi-task attention layer is then designed to dynamically learn the contribution of each high-dimensional representation, ultimately predicting whether the patient has bladder or urethral abnormalities and determining the grade of ureteral reflux.

[0066] Furthermore, the construction and processing of each module in the task of automatically segmenting and diagnosing multiple VCUG images of a single child are as follows:

[0067] 1. Preprocessing of input VCUG images

[0068] Preprocessing experimental data is a crucial step in improving data quality. It involves standardizing images and reducing noise and interference through simple operations. In this study, we first cropped all images to remove edge annotations. Then, we filled the image edges with black to create squares of equal length and width. Next, we uniformly scaled the images to 224×224 pixels to provide a consistent input size for the machine learning model. This padding was chosen because the width and length of the ureter in VCUG images are important for VCUG image classification; padding to squares before scaling effectively avoids distortion of the ureter's shape. Finally, we normalized the pixel values ​​to the 0-1 range and standardized the images by channel (subtracting the mean and dividing by the variance), which accelerates model convergence.

[0069] Data augmentation is a common method to reduce overfitting and improve classification performance on small datasets. During training, random data augmentation operations are performed on each image, enabling the machine learning model to learn more generalized features and improving its robustness. In this study, we performed affine transformations, perspective transformations, and irregular deformations on each image; preferably, we randomly rotated each image by 0–15°.

[0070] 2. Segmentation and Diagnosis Model Based on a Single VCUG Image

[0071] We propose a unified transformer-based architecture that performs both multi-label classification and semantic segmentation, aiming to improve overall performance through the complementarity of these tasks. To stabilize training and improve feature representations, we initialize the model on the ImageNet dataset using pre-trained ViT-Base through masked self-supervised learning. This pre-trained initialization accelerates model convergence and improves performance on both classification and segmentation tasks.

[0072] Our model first encodes the input multi-label image into classification tokens and patch tokens. After processing by a Transformer encoder, the classification tokens are fed into a fully connected layer to perform a multi-label classification task. This classification task is divided into four sub-tasks: bladder abnormalities, urethral abnormalities, left ureteral reflux grading, and right ureteral reflux grading. To predict the specific label for each sub-task, we designed a fully connected layer that maps the classification head into a 16-dimensional feature vector, where different parts of the vector correspond to specific sub-task categories: T[0:2] for predicting bladder abnormalities, T[2:4] for predicting urethral abnormalities, T[4:10] for predicting left ureteral reflux grading, and T[10:16] for predicting right ureteral reflux grading.

[0073] While training the classification task, we simultaneously designed a masked Transformer module to perform semantic segmentation, specifically for the bladder, ureter, and urethra. To further enhance the interactivity between the classification and segmentation tasks, we added a self-focused module to facilitate communication between them. Semantic information from the classification task helps the segmentation module better locate regions of interest, while the segmentation task, in turn, provides semantic features that aid in classification. This design promotes information sharing between tasks, generating more expressive high-dimensional feature representations.

[0074] Furthermore, an automatic VCUG segmentation and diagnosis model is constructed based on multiple VCUG images of a single patient as follows: After achieving segmentation and diagnosis based on a single VCUG image, we extend the model to instance-level classification, aiming to automatically identify the contribution of each image in the patient's image set and generate the final classification. The process is as follows:

[0075] First, we load the pre-trained weights from the image-level encoder and freeze these parameters, denoted as T_"frozen". We set the maximum number of images per patient to 10. For patients with fewer than 10 images, zero-value image padding ensures a consistent input size. These images are then sequentially fed into the frozen classifier T_"frozen" to extract high-dimensional feature representations. The features from all images are stacked into a matrix, denoted as H∈R^(10×D), where D represents the feature dimension.

[0076] To perform instance-level classification, we introduce a set of K learnable class embeddings, “cls” = [cls_1, ..., cls_K] ∈ R^(K×D), where K = 3, corresponding to the tasks of classifying bladder abnormalities, urethral abnormalities, and ureteral reflux. Each class embedding, along with a high-dimensional feature matrix H of 10 images, is processed by a decoder. Finally, we use three independent classification heads, each mapping a class embedding to a probability distribution of possible disease categories. This structure enables the model to automatically learn the contribution of each image to each disease classification task.

[0077] The training results of the model in a single center are shown in Tables 1 and 2.

[0078] Table 1. Diagnostic performance of the model on a single image on the internal validation dataset.

[0079]

[0080] Table 2. Diagnostic performance of the model on a single patient on the internal validation dataset.

[0081]

[0082] The foregoing description of embodiments of the present invention, through which those skilled in the art are able to implement or use the present invention, will be readily apparent to those skilled in the art. Various modifications to these embodiments will be readily apparent to those skilled in the art. The general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novelty disclosed herein.

Claims

1. A method for automated diagnosis of excretory cystourethrography using deep learning, characterized in that, The method includes the following steps: Step 1: Data Acquisition and Image Selection Images of children undergoing excretory cystourethrography were collected, and images with poor quality were excluded; Step 2, Image Annotation and Dataset Construction: The selected VCUG images are segmented and classified into the bladder, urethra, and areas with vesicoureteral reflux. If there is a discrepancy between the two, the doctor will make the final classification to complete the collection of the dataset. Step 3: Deep learning model construction: An automatic diagnostic model for VCUG images is constructed using deep learning methods. The model consists of two parts: 1) Segmentation and classification based on a single VCUG image: A unified transformer-based architecture is used to handle classification and segmentation tasks simultaneously; the input is a single VCUG image, and the output is a lesion mask and diagnosis; 2) Construct an automatic segmentation and diagnosis model for VCUG images based on multiple VCUG images of a single patient: For each VCUG image, based on the output of the first part of the model, extract the high-dimensional feature representation of each image, and superimpose the high-dimensional feature representation of each image with the disease-related representation into a matrix; then design a multi-task attention layer to dynamically learn the contribution of each high-dimensional representation, and finally predict whether the patient has bladder or urethral abnormalities and determine the grade of ureteral reflux; Step 4: Testing the deep learning model: Multiple VCUG images from a new single patient are input, and image preprocessing is performed first. Then, the standardized VCUG images are input into the constructed automatic diagnostic model. The model automatically segments the bladder, urethra, and ureter regions and diagnoses lesions in the segmented regions.

2. The automatic diagnostic method according to claim 1, characterized in that, The data acquisition and image screening include: collecting a sufficient number of VCUG images; image annotation and dataset construction include: doctors segmenting the bladder, urethra, and ureter regions in a single VCUG image and making a diagnosis, the diagnosis results including whether the bladder and urethra are abnormal, and the degree of vesicoureteral reflux; for multiple VCUG images of a single person, a final diagnosis is made after comprehensive judgment based on the results of each VCUG image, the diagnosis results including whether the bladder and urethra are abnormal, and the degree of vesicoureteral reflux.

3. The automatic diagnostic method according to claim 1, characterized in that, Step 2 of step 3 includes: using an automated diagnostic network model for excretory vesicoureterostomy, firstly segmenting each obtained VCUG image into the bladder, urethra, and ureter, and then classifying the obtained segmented regions; next, for multiple VCUG images of a single person, the model integrates the segmentation and diagnostic results of each VCUG image to calculate the contribution, and finally makes a final diagnosis, including whether the bladder and urethra are abnormal, and the grade of vesicoureteral reflux.

4. The automatic diagnostic method according to claim 1, characterized in that, The preprocessing of each input VCUG image includes: cropping and removing edge annotations, filling the image edges with black to make the image shape a square, then uniformly scaling the image to a size of 224×224 pixels, and finally normalizing the pixel values ​​of the image to the range of 0~1, and standardizing the image according to the channels, namely affine transformation, perspective transformation, irregular deformation, and adding random noise.

5. The automatic diagnostic method according to claim 4, characterized in that, When normalizing images by channel, each image is randomly rotated by 0 to 15 degrees.

6. The automatic diagnostic method according to claim 1, characterized in that, The automated diagnostic method model for excretory cystourethrography includes: 1) A dataset that collects VCUG images with segmentation and diagnostic results as training samples; 2) Data augmentation techniques are used on the collected training sample dataset to increase the number and variety of images by transforming them, thereby improving the generalization ability of the network model and its robustness to image segmentation. 3) Deep learning feature extraction networks; 4) A model for the automatic diagnosis of excretory cystourethrography.

7. The automatic diagnostic method according to claim 6, characterized in that, The deep learning models involved in excretory cystourethrography are all based on convolutional neural networks for feature extraction. The processed VCUG image is used as input, and the feature extraction is performed through convolutional layers, pooling layers and activation function layers to finally obtain a feature map with complex semantic information. The automatic diagnostic model for excretory cystourethrography includes: unfolding the feature map of the VCUG image to obtain the corresponding feature vector, performing feature fusion through a fully connected layer, calculating the probability of classifying the image into each level through the softmax function, and finally comparing the magnitude of the probabilities to determine the level.

8. An automatic diagnostic system employing the method described in any one of claims 1-7, characterized in that, The system includes: The image acquisition module is used to acquire images of children undergoing excretory cystourethrography and to filter the images, excluding those with poor image quality. The image annotation module is used to segment and classify the bladder, urethra, and ureter regions of the selected VCUG images and generate an annotated dataset. The deep learning model module is used to build an automatic diagnostic model based on deep learning. The model includes a segmentation and classification module based on a single VCUG image, and a comprehensive diagnostic module based on multiple VCUG images of a single person. The model dynamically learns the contribution of each image through a multi-task attention layer and outputs diagnostic results, including whether the bladder and urethra are abnormal and the degree of vesicoureteral reflux. The data preprocessing module is used to preprocess the input VCUG image, including cropping, filling, scaling and normalization operations; The automatic diagnosis module receives multiple VCUG images from patients, performs image segmentation and diagnosis, and generates segmentation and lesion diagnosis results for the bladder, urethra and ureter regions by analyzing the high-dimensional features of multiple images.

9. The automatic diagnostic system according to claim 8, characterized in that, The deep learning model module is initialized using a ViT-base model pre-trained on the ImageNet dataset. It utilizes mask self-supervised learning technology to improve the system's feature extraction capability for medical images and accelerate the model training process.

10. The automatic diagnostic system according to claim 8, characterized in that, The system also includes a feature fusion module, which fuses the feature vectors of each image through a fully connected layer, calculates the probability of classifying the image into each level using the softmax function, and finally makes a diagnostic level decision based on the probability magnitude.

Images