Image Features Research Articles

Research on carotid artery plaque anomaly detection algorithm based on ultrasound images

Carotid artery plaque is a key factor in stroke and other cardiovascular diseases. Accurate detection and localization of carotid artery plaque are essential for early prevention and treatment of diseases. However, current carotid artery ultrasound image anomaly detection algorithms face several challenges, such as scarcity of anomaly data in carotid arteries and traditional convolutional neural networks (CNNs) overlooking long-distance dependencies in image processing. To address these issues, we propose an anomaly detection algorithm for carotid artery plaques based on ultrasound images. The algorithm innovatively introduces an anomaly sample pair generation method to increase dataset diversity. Moreover, it employs an improved adaptive recursive gating pyramid pooling module to extract image features. This module significantly enhances the model’s capacity for high-order spatial interactions and adaptive feature fusion, thereby greatly improving the neural network’s feature extraction ability. The algorithm uses a Sigmoid layer to map each pixel’s feature vector to a probability distribution between 0 and 1, and anomalies are detected through probability threshold binarization. Experimental results show that our algorithm’s AUROC index reached 90.7% on a carotid artery dataset, improving by 2.1% compared to the FPI method. This research is expected to provide robust support for the early prevention and treatment of cardiovascular diseases.

A novel hybrid deep learning IChOA-CNN-LSTM model for modality-enriched and multilingual emotion recognition in social media.

In the rapidly evolving field of artificial intelligence, the importance of multimodal sentiment analysis has never been more evident, especially amid the ongoing COVID-19 pandemic. Our research addresses the critical need to understand public sentiment across various dimensions of this crisis by integrating data from multiple modalities, such as text, images, audio, and videos sourced from platforms like Twitter. Conventional methods, which primarily focus on text analysis, often fall short in capturing the nuanced intricacies of emotional states, necessitating a more comprehensive approach. To tackle this challenge, our proposed framework introduces a novel hybrid model, IChOA-CNN-LSTM, which leverages Convolutional Neural Networks (CNNs) for precise image feature extraction, Long Short-Term Memory (LSTM) networks for sequential data analysis, and an Improved Chimp Optimization Algorithm for effective feature fusion. Remarkably, our model achieves an impressive accuracy rate of 97.8%, outperforming existing approaches in the field. Additionally, by integrating the GeoCoV19 dataset, we facilitate a comprehensive analysis that spans linguistic and geographical boundaries, enriching our understanding of global pandemic discourse and providing critical insights for informed decision-making in public health crises. Through this holistic approach and innovative techniques, our research significantly advances multimodal sentiment analysis, offering a robust framework for deciphering the complex interplay of emotions during unprecedented global challenges like the COVID-19 pandemic.

X-ray Bone Image Processing Based on Improved Densenet

The application of convolutional neural network has been gradually introduced in the field of X-ray bone images. At present, there is less research on methods to automatically detect abnormal parts of the skeleton. The method improves on the Densenet network by adjusting the network structure, adding a convolutional block attention module (CBAM) attention mechanism to the original Densenlayer, fusing spatial attention and channel attention to suppress unnecessary features and strengthen the network‘s capability to extract image features, and optimizing the Transition block by fusing two pooling strategies, average and maximum, to increase the model’s anti-interference capability. The data collected after the experiment show that the improved Densenet has increased accuracy in the detection of skeletal abnormalities at all sites compared with the traditional network. The average accuracy is 1–5% higher than the benchmark method, the highest elbow accuracy reached 77.62%, the lowest hand accuracy also reached 64.28% and the area under the receiver operating characteristic curve is increased by 3–7%, the highest reached 80.83%.

Retina-Inspired Lightweight Spiking Convolutional Neural Network for Single-Image Dehazing.

Suspended particles in hazy medium absorb and scatter light, severely degrading imaging quality. Numerous single-image dehazing methods have been proposed to reconstruct clear images from hazy ones. However, most of them focus on increasing depth and width to improve dehazing performance, which incurs high computation and energy costs. To address this issue, we propose a lightweight spiking convolutional neural network (CNN) referred to as retina-inspired spiking CNN (RI-SCNN) for the reconstruction of hazy images. Unlike conventional dehazing techniques, first, our proposed network simulates the hierarchical structure and cellular function of the retina and devises five network modules to efficiently encode and extract image features through ON and OFF roads. Furthermore, the linear reconstruction mechanism is introduced to integrate the outputs from different roads, adaptively preserving regions with optimal details and constructing a comprehensive visual representation. Finally, by the transformed atmospheric scattering formula, our network can generate the dehazy image. Incorporating the microscale spiking mechanism of the brain, the entire network leverages discrete binary spike trains for information encoding and transmission, directly trained by spiking surrogate gradient learning on integrate-and-fire (IF) neurons. Experimental results demonstrate the superiority of the proposed RI-SCNN in terms of quantitative dehazing performance, qualitative visual effect, energy efficiency, and run speed. Considering its lightweight architecture with ultralow computation and energy costs, the network is encouraged to be deployed in the visual sensor hardware to improve overall performance.

SMLS-YOLO: an extremely lightweight pathological myopia instance segmentation method.

Pathological myopia is a major cause of blindness among people under 50 years old and can result in severe vision loss in extreme cases. Currently, its detection primarily relies on manual methods, which are slow and heavily dependent on the expertise of physicians, making them impractical for large-scale screening. To tackle these challenges, we propose SMLS-YOLO, an instance segmentation method based on YOLOv8n-seg. Designed for efficiency in large-scale screenings, SMLS-YOLO employs an extremely lightweight model. First, StarNet is introduced as the backbone of SMLS-YOLO to extract image features. Subsequently, the StarBlock from StarNet is utilized to enhance the C2f, resulting in the creation of the C2f-Star feature extraction module. Furthermore, shared convolution and scale reduction strategies are employed to optimize the segmentation head for a more lightweight design. Lastly, the model incorporates the Multi-Head Self-Attention (MHSA) mechanism following the backbone to further refine the feature extraction process. Experimental results on the pathological myopia dataset demonstrate that SMLS-YOLO outperforms the baseline YOLOv8n-seg by reducing model parameters by 46.9%, increasing Box mAP@0.5 by 2.4%, and enhancing Mask mAP@0.5 by 4%. Furthermore, when compared to other advanced instance segmentation and semantic segmentation algorithms, SMLS-YOLO also maintains a leading position, suggesting that SMLS-YOLO has promising applications in the segmentation of pathological myopia images.

Joint contrastive self-supervised learning and weak-orthogonal product quantization for fast image retrieval

Hashing and quantization methods based on deep learning are being widely used in large-scale image retrieval. Most methods use traditional supervised models to train models. Although these methods achieve excellent performance, they rely too much on expensive label information and do not fully utilize data resources. These methods overly encourage binary codes to retain original information, which may lead to models building too much useless information at the expense of discriminative information that is more important for retrieval. To address these issues, we propose a self-supervised method that does not require labeled information, called Contrastive Self-Supervised Weak-Orthogonal Product Quantization (CSWPQ). The method is trained in a label-free self-supervised manner to reduce the need for labels by maximizing the similarity of different views of the same image and minimizing the similarity of different image views, which not only improves the generalization ability of the model but also enhances the robustness. We introduce a quantified contrastive learning method that is different from traditional contrastive learning to learn more discriminative image representations by constructing quantified image features of different views of the image. In addition, in order to reduce the quantization error caused by product quantization, we impose a weak-orthogonal constraint to increase the diversity of representations, reduce the information loss and computational overhead in the quantization process, and improve the computational speed. Extensive experiments on CIFAR-10, NUS-WIDE and FLICKR25K datasets show that our proposed method achieves better performance.

A modality separation approach for facial sketch synthesis

AbstractThe technology for face‐to‐sketch synthesis transforms optical face images into a sketch‐style format. However, traditional style losses are insufficient to discern the modal differences between optical and sketch domain images, leading to unclear images. At the same time, generated images lack clarity due to traditional approaches' disregard for high‐frequency texture. To address these issues, a modality separation approach for facial sketch synthesis is proposed. First, a modality separation structure is proposed, using a quicksort algorithm to merge features of optical and sketch images as target modality (positive samples), ensuring the generated images' feature distribution matches real sketches. By controlling the Euclidean distance between generated images (anchors) and both target and filtered modality (positive and negative samples), irrelevant information is effectively filtered out. Next, an edge‐promoting module feeds processed blurry sketch images into the discriminator to enhance robustness. Lastly, a detail optimization module uses Laplacian filtering to extract high‐frequency texture from optical face images for local enhancement. Experimental validation on CUHK, AR, and XM2VTS datasets shows that this method outperforms mainstream sketch face synthesis methods in terms of Fréchet inception distance and learned perceptual image patch similarity, producing more realistic and natural images with richer texture details.

Traditional landscape painting and art image restoration methods based on structural information guidance

Abstract In the field of traditional landscape painting and art image restoration, traditional restoration methods have gradually revealed limitations with the development of society and technological progress. In order to enhance the restoration effects of Chinese landscape paintings, an innovative image restoration algorithm is designed in this research, combining edge restoration with generative adversarial networks (GANs). Simultaneously, a novel image restoration model with embedded multi-scale attention dilated convolution is proposed to enhance the modeling capability for details and textures in landscape paintings. To better preserve the structural features of artistic images, a structural information-guided art image restoration model is introduced. The introduction of adversarial networks into the repair model can improve the repair effect. The art image repair model adds a multi-scale attention mechanism to handle more complex works of art. The research results show that the image detection model improves by 0.20, 0.07, and 0.06 in the Spearman rank correlation coefficient, Pearson correlation coefficient, and peak signal-to-noise ratio (PSNR), respectively, compared to other models. The proposed method outperforms mean filtering, wavelet denoising, and median filtering algorithms by 6.3, 9.1, and 15.8 dB in PSNR and by 0.06, 0.12, and 0.11 in structural similarity index. In the image restoration task, the structural similarity and information entropy indicators of the research model increase by approximately 9.3 and 3%, respectively. The image restoration method proposed in this study is beneficial for preserving and restoring precious cultural heritage, especially traditional Chinese landscape paintings, providing new technological means for cultural relic restoration.

SwinDenoising: A Local and Global Feature Fusion Algorithm for Infrared Image Denoising

Infrared image denoising is a critical task in various applications, yet existing methods often struggle with preserving fine details and managing complex noise patterns, particularly under high noise levels. To address these limitations, this paper proposes a novel denoising method based on the Swin Transformer architecture, named SwinDenoising. This method leverages the powerful feature extraction capabilities of Swin Transformers to capture both local and global image features, thereby enhancing the denoising process. The proposed SwinDenoising method was tested on the FLIR and KAIST infrared image datasets, where it demonstrated superior performance compared to state-of-the-art methods. Specifically, SwinDenoising achieved a PSNR improvement of up to 2.5 dB and an SSIM increase of 0.04 under high levels of Gaussian noise (50 dB), and a PSNR increase of 2.0 dB with an SSIM improvement of 0.03 under Poisson noise (λ = 100). These results highlight the method’s effectiveness in maintaining image quality while significantly reducing noise, making it a robust solution for infrared image denoising.

TAMC: Textual Alignment and Masked Consistency for Open-Vocabulary 3D Scene Understanding.

Three-dimensional (3D) Scene Understanding achieves environmental perception by extracting and analyzing point cloud data with wide applications including virtual reality, robotics, etc. Previous methods align the 2D image feature from a pre-trained CLIP model and the 3D point cloud feature for the open vocabulary scene understanding ability. We believe that existing methods have the following two deficiencies: (1) the 3D feature extraction process ignores the challenges of real scenarios, i.e., point cloud data are very sparse and even incomplete; (2) the training stage lacks direct text supervision, leading to inconsistency with the inference stage. To address the first issue, we employ a Masked Consistency training policy. Specifically, during the alignment of 3D and 2D features, we mask some 3D features to force the model to understand the entire scene using only partial 3D features. For the second issue, we generate pseudo-text labels and align them with the 3D features during the training process. In particular, we first generate a description for each 2D image belonging to the same 3D scene and then use a summarization model to fuse these descriptions into a single description of the scene. Subsequently, we align 2D-3D features and 3D-text features simultaneously during training. Massive experiments demonstrate the effectiveness of our method, outperforming state-of-the-art approaches.

Cross-attention guided loss-based deep dual-branch fusion network for liver tumor classification

Recently, convolutional neural networks (CNNs) and multiple instance learning (MIL) methods have been successfully applied to MRI images. However, CNNs directly utilize the whole image as the model input and the downsampling strategy (like max or mean pooling) to reduce the size of the feature map, thereby possibly neglecting some local details. And MIL methods learn instance-level or local features without considering spatial information. To overcome these issues, in this paper, we propose a novel cross-attention guided loss-based dual-branch framework (LCA-DB) to leverage spatial and local image information simultaneously, which is composed of an image-based attention network (IA-Net), a patch-based attention network (PA-Net) and a cross-attention module (CA). Specifically, IA-Net directly learns image features with loss-based attention to mine significant regions, meanwhile, PA-Net captures patch-specific representations to extract crucial patches related to the tumor. Additionally, the cross-attention module is designed to integrate patch-level features by using attention weights generated from each other, thereby assisting them in mining supplement region information and enhancing the interactive collaboration of the two branches. Moreover, we employ an attention similarity loss to further reduce the semantic inconsistency of attention weights obtained from the two branches. Finally, extensive experiments on three liver tumor classification tasks demonstrate the effectiveness of the proposed framework, e.g., on the LLD-MMRI–7, our method achieves 69.2%, 65.9% and 88.5% on the seven-class liver tumor classification tasks in terms of accuracy, F1 score and AUC, with the superior classification and interpretation performance over recent state-of-the-art methods. The source code of LCA-DB is available at https://github.com/Wangrui-berry/Cross-attention.

Global and local feature extraction based on convolutional neural network residual learning for MR image denoising.

Given the different noise distribution information of global and local magnetic resonance (MR) images, this study aims to extend the current work on convolutional neural networks that preserve global structure and local details in MR image denoising tasks. This study proposed a parallel and serial network for denoising 3D MR images, called 3D-PSNet. We use the residual depthwise separable convolution block to learn the local information of the feature map, reduce the network parameters, and thus improve the training speed and parameter efficiency. In addition, we consider the feature extraction of the global image and utilize residual dilated convolution to process the feature map to expand the receptive field of the network and avoid the loss of global information. Finally, we combine both of them to form a parallel network. What's more, we integrate reinforced residual convolution blocks with dense connections to form serial network branches, which can remove redundant information and refine features to further obtain accurate noise information. The peak signal-to-noise ratio, structural similarity index measure, and root mean square error metrics of 3D-PSNet are as high as 47.79%, 99.81%, and 0.40%, respectively, achieving competitive denoising effect on three public datasets. The ablation experiments demonstrated the effectiveness of all the designed modules regarding all the evaluated metrics in both datasets. The proposed 3D-PSNet takes advantage of multi-scale receptive fields, local feature extraction and residual dense connections to more effectively restore the global structure and local fine features in MR images, and is expected to help doctors quickly and accurately diagnose patients' conditions.

Machine learning for automated classification of lung collagen in a urethane-induced lung injury mouse model.

Dysregulation of lung tissue collagen level plays a vital role in understanding how lung diseases progress. However, traditional scoring methods rely on manual histopathological examination introducing subjectivity and inconsistency into the assessment process. These methods are further hampered by inter-observer variability, lack of quantification, and their time-consuming nature. To mitigate these drawbacks, we propose a machine learning-driven framework for automated scoring of lung collagen content. Our study begins with the collection of a lung slide image dataset from adult female mice using second harmonic generation (SHG) microscopy. In our proposed approach, first, we manually extracted features based on the 46 statistical parameters of fibrillar collagen. Subsequently, we pre-processed the images and utilized a pre-trained VGG16 model to uncover hidden features from pre-processed images. We then combined both image and statistical features to train various machine learning and deep neural network models for classification tasks. We employed advanced unsupervised techniques like K-means, principal component analysis (PCA), t-distributed stochastic neighbour embedding (t-SNE), and uniform manifold approximation and projection (UMAP) to conduct thorough image analysis for lung collagen content. Also, the evaluation of the trained models using the collagen data includes both binary and multi-label classification to predict lung cancer in a urethane-induced mouse model. Experimental validation of our proposed approach demonstrates promising results. We obtained an average accuracy of 83% and an area under the receiver operating characteristic curve (ROC AUC) values of 0.96 through the use of a support vector machine (SVM) model for binary categorization tasks. For multi-label classification tasks, to quantify the structural alteration of collagen, we attained an average accuracy of 73% and ROC AUC values of 1.0, 0.38, 0.95, and 0.86 for control, baseline, treatment_1, and treatment_2 groups, respectively. Our findings provide significant potential for enhancing diagnostic accuracy, understanding disease mechanisms, and improving clinical practice using machine learning and deep learning models.

Light-Aware Contrastive Learning for Low-Light Image Enhancement

Low-Light Image Enhancement (LLIE) presents challenges due to texture information loss and uneven illumination, which can distort feature distribution and reduce the quality of the enhanced images. However, current deep learning methods for LLIE only use supervised information from clear images to extract low-light image features, while disregarding the negative information in low-light images (i.e., low illumination and noise). To address these challenges, we propose a novel LLIE method, LACR-VAE, by leveraging the negative information and considering the uneven illumination. In particular, a Light-Aware Contrastive Regularization (LACR) based on contrastive learning is designed to exploit information from both clear and low-light images. The LACR aims to align latent variables of enhanced images with clear images, away from those of low-light images. This allows the method to prioritize essential elements for LLIE and minimize noise and lighting variations. Furthermore, considering the uneven illumination with diverse region sizes and shapes, a Region-CAlibrated Module (RCAM) is present to learn local and global illumination relations among image regions, and an Attention-guided Multi-Scale Module (AMSM) is designed to extract multi-scale features that improve the model’s representation capability. Extensive experiments show that our method achieves superior performance than previous works. Specifically, our method yields a significant enhancement in the National Aeronautics and Space Administration (NASA) testset, achieving an improvement of at least 0.99 in PSNR and 0.0409 in SSIM. Codes and datasets are available at https://github.com/csxuwu/LACR-VAE .

A multi-scale deep residual network-based guided wave imaging evaluation method for fatigue crack quantification

Abstract As a promising structural health monitoring (SHM) technology, guided wave (GW) imaging is gaining increasing attention for crack monitoring of aircraft structures. However, actual fatigue crack propagation is a complex dynamically evolving process affected by various variabilities. It is still challenging to accurately track and quantify the dynamic fatigue crack propagation with GW imaging methods. Therefore, in order to achieve more accurate fatigue crack quantification, this paper proposes a multi-scale deep residual network-based GW imaging evaluation method. A convolutional neural network (CNN) is utilized to evaluate the entire pixel distribution of GW imaging maps to fuse damage-related information from multiple GW monitoring paths. By designing multi-scale convolutional kernels and deep residual learning, a robust quantitative image feature extraction is ensured with the dynamic evolution process of fatigue crack growth and the performance degradation is avoided as the CNN goes deeper, thereby improving the quantification accuracy. The method is validated on a fatigue test of landing gear beams, which are important load-carrying aircraft structural components. The results demonstrate that the proposed method can extract multi-scale crack length-related features and accurately track fatigue crack propagations. For batch specimens, the maximum quantification error is reduced from the original 6.1mm to 1.6mm, marking a significant improvement.

Application of the Semi-Supervised Learning Approach for Pavement Defect Detection.

Road surface quality is essential for driver comfort and safety, making it crucial to monitor pavement conditions and detect defects in real time. However, the diversity of defects and the complexity of ambient conditions make it challenging to develop an effective and robust classification and detection algorithm. In this study, we adopted a semi-supervised learning approach to train ResNet-18 for image feature retrieval and then classification and detection of pavement defects. The resulting feature embedding vectors from image patches were retrieved, concatenated, and randomly sampled to model a multivariate normal distribution based on the only one-class training pavement image dataset. The calibration pavement image dataset was used to determine the defect score threshold based on the receiver operating characteristic curve, with the Mahalanobis distance employed as a metric to evaluate differences between normal and defect pavement images. Finally, a heatmap derived from the defect score map for the testing dataset was overlaid on the original pavement images to provide insight into the network's decisions and guide measures to improve its performance. The results demonstrate that the model's classification accuracy improved from 0.868 to 0.887 using the expanded and augmented pavement image data based on the analysis of heatmaps.

Multimodal contrastive learning for spatial gene expression prediction using histology images.

In recent years, the advent of spatial transcriptomics (ST) technology has unlocked unprecedented opportunities for delving into the complexities of gene expression patterns within intricate biological systems. Despite its transformative potential, the prohibitive cost of ST technology remains a significant barrier to its widespread adoption in large-scale studies. An alternative, more cost-effective strategy involves employing artificial intelligence to predict gene expression levels using readily accessible whole-slide images stained with Hematoxylin and Eosin (H&E). However, existing methods have yet to fully capitalize on multimodal information provided by H&E images and ST data with spatial location. In this paper, we propose mclSTExp, a multimodal contrastive learning with Transformer and Densenet-121 encoder for Spatial Transcriptomics Expression prediction. We conceptualize each spot as a "word", integrating its intrinsic features with spatial context through the self-attention mechanism of a Transformer encoder. This integration is further enriched by incorporating image features via contrastive learning, thereby enhancing the predictive capability of our model. We conducted an extensive evaluation of highly variable genes in two breast cancer datasets and a skin squamous cell carcinoma dataset, and the results demonstrate that mclSTExp exhibits superior performance in predicting spatial gene expression. Moreover, mclSTExp has shown promise in interpreting cancer-specific overexpressed genes, elucidating immune-related genes, and identifying specialized spatial domains annotated by pathologists. Our source code is available at https://github.com/shizhiceng/mclSTExp.

An Innovative Hybrid Model for Automatic Detection of White Blood Cells in Clinical Laboratories.

Background: Microscopic examination of peripheral blood is a standard practice in clinical medicine. Although manual examination is considered the gold standard, it presents several disadvantages, such as interobserver variability, being quite time-consuming, and requiring well-trained professionals. New automatic digital algorithms have been developed to eliminate the disadvantages of manual examination and improve the workload of clinical laboratories. Objectives: Regular analysis of peripheral blood cells and careful interpretation of their results are critical for protecting individual health and early diagnosis of diseases. Because many diseases can occur due to this, this study aims to detect white blood cells automatically. Methods: A hybrid model has been developed for this purpose. In the developed model, feature extraction has been performed with MobileNetV2 and EfficientNetb0 architectures. In the next step, the neighborhood component analysis (NCA) method eliminated unnecessary features in the feature maps so that the model could work faster. Then, different features of the same image were combined, and the extracted features were combined to increase the model's performance. Results: The optimized feature map was classified into different classifiers in the last step. The proposed model obtained a competitive accuracy value of 95.6%. Conclusions: The results obtained in the proposed model show that the proposed model can be used in the detection of white blood cells.

DETR-ORD: An Improved DETR Detector for Oriented Remote Sensing Object Detection with Feature Reconstruction and Dynamic Query

Optical remote sensing images often feature high resolution, dense target distribution, and uneven target sizes, while transformer-based detectors like DETR reduce manually designed components, DETR does not support arbitrary-oriented object detection and suffers from high computational costs and slow convergence when handling large sequences of images. Additionally, bipartite graph matching and the limit on the number of queries result in transformer-based detectors performing poorly in scenarios with multiple objects and small object sizes. We propose an improved DETR detector for Oriented remote sensing object detection with Feature Reconstruction and Dynamic Query, termed DETR-ORD. It introduces rotation into the transformer architecture for oriented object detection, reduces computational cost with a hybrid encoder, and includes an IFR (image feature reconstruction) module to address the loss of positional information due to the flattening operation. It also uses ATSS to select auxiliary dynamic training queries for the decoder. This improved DETR-based detector enhances detection performance in challenging oriented optical remote sensing scenarios with similar backbone network parameters. Our approach achieves superior results on most optical remote sensing datasets, such as DOTA-v1.5 (72.07% mAP) and DIOR-R (66.60% mAP), surpassing the baseline detector.

Federated learning via multi-attention guided UNet for thyroid nodule segmentation of ultrasound images

Accurate segmentation of thyroid nodules is essential for early screening and diagnosis, but it can be challenging due to the nodules' varying sizes and positions. To address this issue, we propose a multi-attention guided UNet (MAUNet) for thyroid nodule segmentation. We use a multi-scale cross attention (MSCA) module for initial image feature extraction. By integrating interactions between features at different scales, the impact of thyroid nodule shape and size on the segmentation results has been reduced. Additionally, we incorporate a dual attention (DA) module into the skip-connection step of the UNet network, which promotes information exchange and fusion between the encoder and decoder. To test the model's robustness and effectiveness, we conduct the extensive experiments on multi-center ultrasound images provided by 17 local hospitals. The model is trained using the federal learning mechanism to ensure privacy protection. The experimental results show that the Dice scores of the model on the data sets from the three centers are 0.908, 0.912 and 0.887, respectively. Compared to existing methods, our method demonstrates higher generalization ability on multi-center datasets and achieves better segmentation results.