Blood vessel and focus multitask segmentation method based on ultra-wide-angle fundus image
Through the multi-task semi-supervised learning algorithm and cross-level non-local graph module with weight control mechanism, combined with encoder and decoder, the combined segmentation of blood vessels and lesion areas in ultra-wide-angle fundus images is realized, solving the information interaction of single-task learning Inadequacy and labeling problems improve segmentation accuracy and reliability.
Patent Information
- Application Number
- CN202411234100.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-04
- Publication Date
- 2024-11-08
- Estimated Expiration
- 2044-09-04
AI Technical Summary
The existing ultra-wide-angle fundus image processing methods mainly focus on single-task learning, and cannot fully utilize multiple information in the image, which increases training complexity, and is difficult to obtain labels, resulting in difficult to achieve fine and accurate segmentation results.
Using a multi-task semi-supervised learning algorithm based on weight control mechanism, the cross-level non-local graph module is fused with high-level and low-level features, combined with encoder and decoder for feature extraction and reconstruction, and a small amount of labeled data and a large amount of unlabeled data are used to realize blood vessels. Combined segmentation with lesion area.
It effectively reduces the labeling workload, improves the accuracy and reliability of vascular and lesion segmentation, outputs fine and accurate segmentation results, and enhances the model's ability to extract image detail features.
Smart Images

Figure CN118918128A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to a method for segmenting an ultra-wide-angle fundus image, and in particular to a multi-task segmentation method for blood vessels and lesions based on an ultra-wide-angle fundus image, and belongs to the field of image processing. Background Art
[0002] As the only vascular tissue in the human body that can be directly observed under non-invasive conditions, retinal blood vessels can improve the convenience and safety of patient diagnosis. By studying and analyzing retinal blood vessels, we can better understand and identify the early symptoms of vascular diseases, thereby achieving early screening and diagnosis. In addition, lesion detection and identification of eye diseases are also crucial. Lesions such as hemorrhage, exudation, hard exudate and macular degeneration may exist in fundus images. Accurate segmentation and analysis of these lesions not only helps to understand the progression of the disease, but also helps to develop more effective treatment plans and monitor the effectiveness of treatment.
[0003] Ultra-Wide-Field (UWF) fundus images have been widely used in the diagnosis and treatment of ophthalmic diseases due to their extremely wide field of view. Unlike other ophthalmic diagnostic technologies, UWF is an advanced non-invasive and non-contact ophthalmic imaging technology. The imaging range of a single image can reach 180 to 200 degrees, covering about 82% of the retinal area, which largely meets the needs of clinicians for the diagnosis of ophthalmic diseases. Clinical practice has shown that the use of computer-aided technology to automatically detect and analyze retinal vascular features and lesion areas in UWF images not only improves the accuracy and efficiency of diagnosis, but also provides important support for the early detection and monitoring of ophthalmic diseases. In addition, UWF technology performs well in capturing fundus details and can identify complex lesions such as retinal detachment, diabetic retinopathy and macular degeneration, providing valuable information for the development of personalized treatment plans. Overall, UWF fundus images show great application potential and development prospects in ophthalmic clinical practice.
[0004] It is crucial to efficiently and accurately extract vascular structure information and lesion area features from UWF images. However, existing methods mainly focus on single-task learning, that is, only extracting vascular structure information or only segmenting lesion areas. This method suppresses the interactivity of information to a certain extent, fails to make full use of the various information in the image, and increases the complexity of training; in addition, the annotation of UWF images is difficult to obtain and is very labor-intensive and material-intensive; finally, how to output detailed and accurate segmentation results is crucial for the diagnosis of fundus diseases. In order to solve these problems, the present invention proposes an end-to-end multi-task semi-supervised learning algorithm, which aims to jointly segment blood vessels and lesion areas in UWF images. By utilizing a small amount of labeled data and a large amount of unlabeled data, while reducing the workload of labeling, high accuracy and reliability are maintained. Summary of the invention
[0005] In view of the deficiencies in the prior art, the present invention develops a multi-task segmentation method for blood vessels and lesions based on ultra-wide-angle fundus images, which can effectively fuse high-level features and low-level features of the image to complete multi-task segmentation of blood vessels and lesions, share weights between the two tasks, enhance the model's ability to extract image detail features, and maintain high blood vessel and lesion segmentation accuracy and reliability while reducing the annotation workload.
[0006] In order to achieve the above object, the present invention is implemented through the following technical solutions: A multi-task segmentation method for blood vessels and lesions based on ultra-wide-angle fundus images, comprising the following steps: S1. Acquire an ultra-wide-angle UWF fundus image dataset, preprocess the images in the UWF fundus image dataset, and divide the dataset into a training set and a test set, wherein the UWF fundus image dataset includes labeled data and unlabeled data; S2. Construct a multi-task semi-supervised learning network based on a weight control mechanism, wherein the multi-task semi-supervised learning network based on a weight control mechanism includes an encoder, a decoder, a cross-hierarchical non-local graph module, and a loss weight control mechanism. The multi-task semi-supervised learning network based on the weight control mechanism is trained using a training set. The images in the training set are subjected to feature extraction by the encoder part of the multi-task semi-supervised learning network based on the weight control mechanism to obtain low-level features. and advanced features ; S3. The low-level features and advanced features Input into the decoder part of the multi-task semi-supervised learning network based on the weight control mechanism for feature decoding and feature reconstruction, so as to obtain a vascular structure segmentation image and a lesion area segmentation image; S4. Optimizing the multi-task semi-supervised learning network based on the weight control mechanism by training the combined loss function and the total loss to obtain a multi-task segmentation model of blood vessels and lesions based on ultra-wide-angle fundus images; S5. Input the data in the test set into the multi-task segmentation model of blood vessels and lesions based on ultra-wide-angle fundus images to obtain UWF fundus blood vessel segmentation results and UWF fundus lesion area segmentation results.
[0007] Furthermore, the preprocessing operation in step S1 includes data rotation and contrast enhancement.
[0008] Furthermore, a skip connection is used between the encoder and the decoder, and a cross-hierarchical non-local graph module is added to the skip connection. Step S3 specifically includes: The skip connection connects the low-level features through the convolutional layers in the cross-level non-local graph module. and advanced features Decomposed into overlapping patches, the overlapping patches are passed to three parallel fully connected layers to calculate the query Q, key K and value V, where the high-level features are As keys and values, low-level features As a query, low-level features are made Matching advanced features , by calculating the similarity between the key and the query, we get the association matrix, and then we get the cross attention result based on the association matrix and the value , the calculation formula is expressed as: Where n represents the dimension of feature embedding. Q represents query, K represents key, and V represents value. The cross-attention results After the noise reduction process is performed by the feedforward module, the cross-attention result with noise eliminated is obtained. , the calculation formula is expressed as: in, represents the affinity matrix, represents the depth convolutional layer, and represents two linear fully connected mapping layers, Represents the ReLU activation function.
[0009] Furthermore, the combined loss function and the total loss in step S4 specifically include: The labeled data and unlabeled data in the UWF fundus image dataset are respectively denoted as U and J, U' is the enhanced data of the labeled data U, and J' is the enhanced data of the unlabeled data J; is the label of U, is the “guessed” label of the unlabeled data J, is the prediction result of labeled data, For the prediction results of unlabeled data, the combined loss function formula is expressed as follows: , , , in, represents the cross entropy loss function, Represents the number of categories. Total loss By labeling data Cross entropy loss and unlabeled data The square loss is composed of, Represents the loss weight coefficient used to control the unlabeled data; The multi-task loss in the training process of the multi-task semi-supervised learning network based on the weight control mechanism consists of a vascular structure loss and a lesion area loss, wherein the vascular structure loss includes a labeled data loss and an unlabeled data loss, and the lesion area loss also includes a labeled data loss and an unlabeled data loss; The multi-task loss calculation formula with labeled data is: , in, and They represent the vascular structure loss in the labeled data and the lesion area loss in the labeled data, respectively. and They represent the weight scalar of the vascular structure task with labeled data and the weight scalar of the lesion task with labeled data, respectively. and are two labels corresponding to the images with labeled data, Indicates the probability that the labeled data belongs to the vascular structure area, , Indicates the probability that the labeled data belongs to the lesion area, , by and Multiply, and Multiplying together, we get and Used to adjust the weight parameters between tasks; The unlabeled multi-task loss is calculated as: , in, and Respectively represent the vascular structure loss in unlabeled data and the lesion area loss in unlabeled data, and They represent the weight scalar of the vascular structure of the unlabeled data and the weight scalar of the lesion area in the unlabeled data, respectively. and They are the predicted probabilities of the vascular structure task in the labeling task and the predicted probabilities of the lesion task in the labeling task. and Used as soft weights to control the vascular structure loss in unlabeled data and the lesion area loss in unlabeled data; The total loss is calculated as: .
[0010] Furthermore, by using the combined loss function and total loss Under supervision, the training optimization of the multi-task semi-supervised learning network based on the weight control mechanism is completed.
[0011] Furthermore, by connecting the features output by the cross-hierarchical non-local graph model with the features output by the decoder, an upsampling operation is performed, and the Sigmod function is used for nonlinear changes. The output is output through the multi-head CNN of the decoder part, and the vascular structure segmentation image and the lesion area segmentation image are output respectively.
[0012] The present invention also provides a computing device, including a memory configured to store computer executable instructions; and a processor configured to execute a multi-task segmentation method for blood vessels and lesions based on ultra-wide-angle fundus images when the computer executable instructions are executed by the processor.
[0013] The advantages of the present invention are: 1. This paper proposes an end-to-end multi-task semi-supervised learning algorithm for the first time to jointly segment blood vessels and lesion areas in UWF images. Through the weight control mechanism, the quantitative analysis of retinal structural indicators is effectively performed, providing a novel and effective way for the diagnosis of fundus diseases.
[0014] 2. This paper proposes a cross-level non-local graph module for the first time, which fuses high-level and low-level features in the decoder stage and adopts a cross-level non-local graph module (CNGM). This module can capture the intrinsic connection between different features while effectively restoring fine-grained details, improving the performance of the network in the final segmentation task.
[0015] 3. The present invention designs a multi-head CNN network to achieve prediction of two tasks, namely, blood vessels and lesions. By using the same decoder, the present invention achieves weight sharing between the two tasks, strengthens the model's ability to extract image detail features, and outputs fine and accurate segmentation results. Two weight control strategies are used for labeled and unlabeled data streams, respectively, so that the blood vessel segmentation task branch and the lesion area segmentation task branch can simultaneously use weak data sets for learning.
[0016] 4. The present invention proposes a multi-task loss function: In the semi-supervised multi-task loss, the loss of labeled data and unlabeled data is comprehensively considered. Through cross entropy loss and square loss, high accuracy and reliability are maintained, and the annotation workload is reduced. BRIEF DESCRIPTION OF THE DRAWINGS
[0017] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used together with the present embodiments to explain the present invention and do not constitute a limitation of the present invention.
[0018] Figure 1 The present invention is a flow chart of the multi-task segmentation method of blood vessels and lesions based on ultra-wide-angle fundus images.
[0019] Figure 2 It is a multi-task semi-supervised learning network based on the weight control mechanism of the present invention.
[0020] Figure 3 It is the cross-hierarchical non-local graph module of the present invention.
[0021] Figure 4 This is the UWF image multi-task segmentation result of the present invention. DETAILED DESCRIPTION
[0022] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0023] Example 1 In this embodiment, Figure 1 As shown, the present invention provides a multi-task segmentation method for blood vessels and lesions based on ultra-wide-angle fundus images, and the specific steps include: S1. Acquire an ultra-wide-angle UWF fundus image dataset, perform data rotation and contrast enhancement on the images in the UWF fundus image dataset, and divide the dataset into a training set and a test set, wherein the UWF fundus image dataset includes labeled data and unlabeled data; S2. Figure 2 As shown, a multi-task semi-supervised learning network based on a weight control mechanism is constructed, wherein the multi-task semi-supervised learning network based on the weight control mechanism includes an encoder, a decoder, a cross-hierarchical non-local graph module (CNGM) and a loss weight control mechanism. A training set is used to train the multi-task semi-supervised learning network based on the weight control mechanism. The images in the training set are subjected to feature extraction by the encoder part of the multi-task semi-supervised learning network based on the weight control mechanism to obtain low-level features. and advanced features ; S3. The low-level features and advanced features The image is input into the decoder part of the multi-task semi-supervised learning network based on the weight control mechanism for feature decoding and feature reconstruction to obtain the vascular structure segmentation image Output. vessel And lesion area segmentation image Output focal ; Specifically, a skip connection is used between the encoder and the decoder, and a cross-hierarchical non-local graph module is added to the skip connection, such as Figure 3 As shown, the skip connection transfers the low-level features through the convolutional layers in the cross-level non-local graph module. and advanced features Decomposed into overlapping patches, the overlapping patches are passed to three parallel fully connected layers Linear to calculate the query Q, key K and value V, where the high-level features are added to the cross-attention module Cross-Attention. As keys and values, low-level features As a query, low-level features are made Matching advanced features , by calculating the similarity between the key and the query, we get the association matrix, and then we get the cross attention result based on the association matrix and the value , the calculation formula is expressed as: Where n represents the dimension of feature embedding. Q represents query, K represents key, and V represents value. The cross-attention results After the noise reduction process is performed by the feed forward module, the cross attention result with noise eliminated is obtained. , the calculation formula is expressed as: in, represents the affinity matrix, represents the depth convolutional layer, and represents two linear fully connected mapping layers, ReLU activation function. Specifically, the features output by the cross-hierarchical non-local graph model are connected with the features output by the decoder, an upsampling operation is performed, and the Sigmod function is used for nonlinear changes. The output is output through the multi-head CNN of the decoder to output the vascular structure segmentation image Output. vessel And lesion area segmentation image Output focal .
[0024] S4. Optimizing the multi-task semi-supervised learning network based on the weight control mechanism by training the combined loss function and the total loss to obtain a multi-task segmentation model of blood vessels and lesions based on ultra-wide-angle fundus images; Specifically, the labeled data and unlabeled data in the UWF fundus image dataset are respectively denoted as U and J, U' is the enhanced data of the labeled data U, and J' is the enhanced data of the unlabeled data J; is the label of U, is the “guessed” label of the unlabeled data J, is the prediction result of labeled data, For the prediction results of unlabeled data, the combined loss function formula is expressed as follows: , , , in, represents the cross entropy loss function, Represents the number of categories. Total loss By labeling data Cross entropy loss and unlabeled data The square loss is composed of, Represents the loss weight coefficient used to control the unlabeled data; The multi-task loss in the training process of the multi-task semi-supervised learning network based on the weight control mechanism consists of a vascular structure loss and a lesion area loss, wherein the vascular structure loss includes a labeled data loss and an unlabeled data loss, and the lesion area loss also includes a labeled data loss and an unlabeled data loss; The multi-task loss calculation formula with labeled data is: , in, and They represent the vascular structure loss in the labeled data and the lesion area loss in the labeled data, respectively. and They represent the weight scalar of the vascular structure task with labeled data and the weight scalar of the lesion task with labeled data, respectively. and are two labels corresponding to the images with labeled data, Indicates the probability that the labeled data belongs to the vascular structure area, , Indicates the probability that the labeled data belongs to the lesion area, , by and Multiply, and Multiplying together, we get and Used to adjust the weight parameters between tasks; The unlabeled multi-task loss is calculated as: , in, and Respectively represent the vascular structure loss in unlabeled data and the lesion area loss in unlabeled data, and They represent the weight scalar of the vascular structure of the unlabeled data and the weight scalar of the lesion area in the unlabeled data, respectively. and They are the predicted probabilities of the vascular structure task in the labeling task and the predicted probabilities of the lesion task in the labeling task. and Used as soft weights to control the vascular structure loss in unlabeled data and the lesion area loss in unlabeled data; The total loss is calculated as: .
[0025] Specifically, by combining the loss function and total loss Under supervision, the training optimization of the multi-task semi-supervised learning network based on the weight control mechanism is completed.
[0026] S5. Input the data in the test set into the multi-task segmentation model of blood vessels and lesions based on ultra-wide-angle fundus images to obtain UWF fundus blood vessel segmentation results and UWF fundus lesion area segmentation results.
[0027] Example 2 In this embodiment, the UWF fundus image dataset is used for experiments, and the effectiveness of the method of the present invention is verified by comparative experiments. We compare the method of the present invention Our with the existing segmentation methods Unet and SwinUnet, and the experimental results are shown in Tables 1 and 2.
[0028] Table 1 Comparison results of blood vessel segmentation between the method of the present invention and the existing method Table 2 Comparison results of lesion segmentation between the method of the present invention and the existing method In this embodiment, Dice coefficient: measures the precision and recall of the model by comparing the overlap between the predicted positive class and the actual positive class; AUC: Area Under the Curve, which represents the area under the ROC curve (Receiver Operating Characteristic curve); FDR: 0 Discovery Rate, which represents the false discovery rate; BACC: Balanced Accuracy, which represents the balanced accuracy; SEN: Sensitivity, which represents the recall rate (Recall) or the true positive rate (1 Positive Rate), which is the proportion of the results that are correctly predicted as positive in all the results that are actually positive.
[0029] In the comparative experiment, we used Dice coefficient, AUC, false discovery rate FDR, balanced accuracy BACC and true positive rate SEN as evaluation indicators. It can be seen from the table that the method of the present invention is superior to other comparative methods in the above indicators, indicating that the method of the present invention has obvious advantages in the multi-task segmentation of blood vessels and lesions in ultra-wide-angle fundus images, and can effectively improve the accuracy of multi-task segmentation of blood vessels and lesions in ultra-wide-angle fundus images.
[0030] Example 3 In this embodiment, Figure 4 As shown in the figure, we use Unet, SwinUnet and the method Our of the present invention to segment the vascular structure and lesion area of the ultra-wide-angle fundus image respectively. The first column in the figure is the original image Original, the first row is the lesion area segmentation results of each method, and the second row is the vascular structure segmentation results of each method. It can be seen that compared with the other two methods, the vascular segmentation and lesion area segmentation results obtained by the method of the present invention are clearer and the segmentation accuracy is higher, which further illustrates the effectiveness of the method of the present invention.
[0031] Example 4 A computing device includes a memory configured to store computer executable instructions; a processor configured to execute a multi-task segmentation method of blood vessels and lesions based on ultra-wide-angle fundus images when the computer executable instructions are executed by the processor.
[0032] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.
Claims
1. A multi-task segmentation method for blood vessels and lesions based on ultra-wide-angle fundus images. It is characterized in that The following steps are involved: S1. Acquire an ultra-wide-angle UWF fundus image dataset, preprocess the images in the UWF fundus image dataset, and divide the dataset into a training set and a test set, wherein the UWF fundus image dataset includes labeled data and unlabeled data; S2. Construct a multi-task semi-supervised learning network based on a weight control mechanism, wherein the multi-task semi-supervised learning network based on a weight control mechanism includes an encoder, a decoder, a cross-hierarchical non-local graph module, and a loss weight control mechanism. The multi-task semi-supervised learning network based on the weight control mechanism is trained using a training set. The images in the training set are subjected to feature extraction by the encoder part of the multi-task semi-supervised learning network based on the weight control mechanism to obtain low-level features. and advanced features ; S3. The low-level features and advanced features Input into the decoder part of the multi-task semi-supervised learning network based on the weight control mechanism for feature decoding and feature reconstruction, so as to obtain a vascular structure segmentation image and a lesion area segmentation image; S4. Optimizing the multi-task semi-supervised learning network based on the weight control mechanism by training the combined loss function and the total loss to obtain a multi-task segmentation model of blood vessels and lesions based on ultra-wide-angle fundus images; S5. Input the data in the test set into the multi-task segmentation model of blood vessels and lesions based on ultra-wide-angle fundus images to obtain UWF fundus blood vessel segmentation results and UWF fundus lesion area segmentation results.
2. According to claim 1, a multi-task segmentation method for blood vessels and lesions based on ultra-wide-angle fundus images, It is characterized in that The preprocessing operation in step S1 includes data rotation and contrast enhancement.
3. The multi-task segmentation method for blood vessels and lesions based on ultra-wide-angle fundus images according to claim 1, It is characterized in that A skip connection is used between the encoder and the decoder, and a cross-hierarchical non-local graph module is added to the skip connection. Step S3 specifically includes: The skip connection connects the low-level features through the convolutional layers in the cross-level non-local graph module. and advanced features Decompose into overlapping patches, which are passed to three parallel fully connected layers to compute the query Q, key K, and value V, where high-level features are As keys and values, low-level features As a query, low-level features are made Matching advanced features , by calculating the similarity between the key and the query, we get the association matrix, and then we get the cross attention result based on the association matrix and the value , the calculation formula is expressed as: Where n represents the dimension of feature embedding, Q represents query, K represents key, and V represents value; The cross-attention results After the noise reduction process is performed by the feedforward module, the cross-attention result with noise eliminated is obtained. , the calculation formula is expressed as: in, represents the affinity matrix, represents the depth convolutional layer, and represents two linear fully connected mapping layers, Represents the ReLU activation function.
4. The multi-task segmentation method for blood vessels and lesions based on ultra-wide-angle fundus images according to claim 1, It is characterized in that The combined loss function and total loss in step S4 specifically include: The labeled data and unlabeled data in the UWF fundus image dataset are respectively denoted as U and J, U' is the enhanced data of the labeled data U, and J' is the enhanced data of the unlabeled data J; is the label of U, is the "guessed" label of the unlabeled data J, is the prediction result of labeled data, For the prediction results of unlabeled data, the combined loss function formula is expressed as follows: , , , in, represents the cross entropy loss function, Indicates the number of categories and the total loss By labeling data Cross entropy loss and unlabeled data The square loss is composed of, Represents the loss weight coefficient used to control the unlabeled data; The multi-task loss in the training process of the multi-task semi-supervised learning network based on the weight control mechanism consists of a vascular structure loss and a lesion area loss, wherein the vascular structure loss includes a labeled data loss and an unlabeled data loss, and the lesion area loss also includes a labeled data loss and an unlabeled data loss; The multi-task loss calculation formula with labeled data is: , in, and They represent the vascular structure loss in the labeled data and the lesion area loss in the labeled data, respectively. and They represent the weight scalar of the vascular structure task with labeled data and the weight scalar of the lesion task with labeled data, respectively. and are two labels corresponding to the images with labeled data, Indicates the probability that the labeled data belongs to the vascular structure area, , Indicates the probability that the labeled data belongs to the lesion area, , by and Multiply, and Multiplying together, we get and Used to adjust the weight parameters between tasks; The unlabeled multi-task loss is calculated as: , in, and Respectively represent the vascular structure loss in unlabeled data and the lesion area loss in unlabeled data, and They represent the weight scalar of the vascular structure of the unlabeled data and the weight scalar of the lesion area in the unlabeled data, respectively. and They are the predicted probabilities of the vascular structure task in the labeling task and the predicted probabilities of the lesion task in the labeling task. and Used as soft weights to control the vascular structure loss in unlabeled data and the lesion area loss in unlabeled data; The total loss is calculated as: ; By combining the loss function and total loss Under supervision, the training optimization of the multi-task semi-supervised learning network based on the weight control mechanism is completed.
5. The multi-task segmentation method for blood vessels and lesions based on ultra-wide-angle fundus images according to claim 3, It is characterized in that By connecting the features output by the cross-hierarchical non-local graph model with the features output by the decoder, an upsampling operation is performed, and the Sigmod function is used for nonlinear changes. The output is output through the multi-head CNN of the decoder part to output the vascular structure segmentation image and the lesion area segmentation image respectively.
6. A computing device, It is characterized in that It includes a memory configured to store computer-executable instructions; and a processor configured to execute a multi-task segmentation method for blood vessels and lesions based on ultra-wide-angle fundus images as described in any one of claims 1 to 5 when the computer-executable instructions are executed by the processor.
Citation Information
Patent Citations
Multi-task multi-classification chest organ segmentation model establishment and segmentation method and system
CN112241966A
Focus segmentation model training and application method based on semi-supervised learning
CN113139974A
Retinal blood vessel segmentation method and device, electronic equipment and storage medium
CN114066884A
Focus segmentation method and device and medium
CN114287879A
Cardiac MRI segmentation method and system based on U-Net and Transform fusion improvement
CN116823850A
Cited By
Deep learning-based quality evaluation method and system for ultra-wide-view eye fundus image
CN119851103A
A Deep Learning-Based Method and System for Quality Assessment of Ultra-Wide Field-of-View Fundus Images
CN119851103B
