Spatial Channel Attention Research Articles

Efficient Generative-Adversarial U-Net for Multi-Organ Medical Image Segmentation

Manual labeling of lesions in medical image analysis presents a significant challenge due to its labor-intensive and inefficient nature, which ultimately strains essential medical resources and impedes the advancement of computer-aided diagnosis. This paper introduces a novel medical image-segmentation framework named Efficient Generative-Adversarial U-Net (EGAUNet), designed to facilitate rapid and accurate multi-organ labeling. To enhance the model’s capability to comprehend spatial information, we propose the Global Spatial-Channel Attention Mechanism (GSCA). This mechanism enables the model to concentrate more effectively on regions of interest. Additionally, we have integrated Efficient Mapping Convolutional Blocks (EMCB) into the feature-learning process, allowing for the extraction of multi-scale spatial information and the adjustment of feature map channels through optimized weight values. Moreover, the proposed framework progressively enhances its performance by utilizing a generative-adversarial learning strategy, which contributes to improvements in segmentation accuracy. Consequently, EGAUNet demonstrates exemplary segmentation performance on public multi-organ datasets while maintaining high efficiency. For instance, in evaluations on the CHAOS T2SPIR dataset, EGAUNet achieves approximately 2% higher performance on the Jaccard metric, 1% higher on the Dice metric, and nearly 3% higher on the precision metric in comparison to advanced networks such as Swin-Unet and TransUnet.

Underwater Object Detection Based on Spatial Pyramid and Channel Attention

In order to solve the problems of low accuracy and high detection delay of conventional object detector in underwater environment, underwater object detection network model based on single-stage target detection is introduced. The specific work is as follows: In order to solve the problem that conventional detectors have low detection accuracy in detecting blurry small targets in underwater scenes due to water quality, a channel spatial attention mechanism is designed, which enables the model to focus more on the feature learning of target objects. This improves the extraction of information within the channel and enhancing the extraction of salient features in cases where the distinction between the foreground and background is not obvious and improves the accuracy on small target in underwater detection. On the basis of the conventional object detector, the spatial pyramid pooling module is designed, which reduces the number of computational parameters required for the extraction of features of the model while maintaining the same receptive field. This improves the inference efficiency of the network, and effectively alleviates the detection delay. The result shows that the improved model can identify underwater targets more accurately, and the detection speed of the model is also improved. The detection accuracy of the model achieved to 80.61% and the FPS achieved to 64.23.

A Full-Scale Shadow Detection Network Based on Multiple Attention Mechanisms for Remote-Sensing Images

Shadows degrade image quality and complicate interpretation, underscoring the importance of accurate shadow detection for many image analysis tasks. However, due to the complex backgrounds and variable shadow characteristics of remote sensing images (RSIs), existing methods often struggle with accurately detecting shadows of various scales and misclassifying dark, non-shaded areas as shadows. To address these issues, we proposed a comprehensive shadow detection network called MAMNet. Firstly, we proposed a multi-scale spatial channel attention fusion module, which extracted multi-scale features incorporating both spatial and channel information, allowing the model to flexibly adapt to shadows of different scales. Secondly, to address the issue of false detection in non-shadow areas, we introduced a criss-cross attention module, enabling non-shadow pixels to be compared with other shadow and non-shadow pixels in the same row and column, learning similar features of pixels in the same category, which improved the classification accuracy of non-shadow pixels. Finally, to address the issue of important information from the other two modules being lost due to continuous upsampling during the decoding phase, we proposed an auxiliary branch module to assist the main branch in decision-making, ensuring that the final output retained the key information from all stages. The experimental results demonstrated that the model outperformed the current state-of-the-art RSI shadow detection method on the aerial imagery dataset for shadow detection (AISD). The model achieved an overall accuracy (OA) of 97.50%, an F1 score of 94.07%, an intersection over union (IOU) of 88.87%, a precision of 95.06%, and a BER of 4.05%, respectively. Additionally, visualization results indicated that our model could effectively detect shadows of various scales while avoiding false detection in non-shadow areas. Therefore, this model offers an efficient solution for shadow detection in aerial imagery.

DSiam-CnK: A CBAM- and KCF-Enabled Deep Siamese Region Proposal Network for Human Tracking in Dynamic and Occluded Scenes.

Despite the accuracy and robustness attained in the field of object tracking, algorithms based on Siamese neural networks often over-rely on information from the initial frame, neglecting necessary updates to the template; furthermore, in prolonged tracking situations, such methodologies encounter challenges in efficiently addressing issues such as complete occlusion or instances where the target exits the frame. To tackle these issues, this study enhances the SiamRPN algorithm by integrating the convolutional block attention module (CBAM), which enhances spatial channel attention. Additionally, it integrates the kernelized correlation filters (KCFs) for enhanced feature template representation. Building on this, we present DSiam-CnK, a Siamese neural network with dynamic template updating capabilities, facilitating adaptive adjustments in tracking strategy. The proposed algorithm is tailored to elevate the Siamese neural network's accuracy and robustness for prolonged tracking, all the while preserving its tracking velocity. In our research, we assessed the performance on the OTB2015, VOT2018, and LaSOT datasets. Our method, when benchmarked against established trackers, including SiamRPN on OTB2015, achieved a success rate of 92.1% and a precision rate of 90.9%. On the VOT2018 dataset, it excelled, with a VOT-A (accuracy) of 46.7%, a VOT-R (robustness) of 135.3%, and a VOT-EAO (expected average overlap) of 26.4%, leading in all categories. On the LaSOT dataset, it achieved a precision of 35.3%, a normalized precision of 34.4%, and a success rate of 39%. The findings demonstrate enhanced precision in tracking performance and a notable increase in robustness with our method.

Forecasting of Local Lightning Using Spatial–Channel-Enhanced Recurrent Convolutional Neural Network

Lightning is a hazardous weather phenomenon, characterized by sudden occurrences and complex local distributions. It poses significant challenges for accurate forecasting, which is crucial for public safety and economic stability. Deep learning methods are often better than traditional numerical weather prediction (NWP) models at capturing the spatiotemporal predictors of lightning events. However, these methods struggle to integrate predictors from diverse data sources, which leads to lower accuracy and interpretability. To address these challenges, the Multi-Scale Spatial–Channel-Enhanced Recurrent Convolutional Neural Network (SCE-RCNN) is proposed to improve forecasting accuracy and timeliness by utilizing multi-source data and enhanced attention mechanisms. The proposed model incorporates a multi-scale spatial–channel attention module and a cross-scale fusion module, which facilitates the integration of data from diverse sources. The multi-scale spatial–channel attention module utilizes a multi-scale convolutional network to extract spatial features at different spatial scales and employs a spatial–channel attention mechanism to focus on the most relevant regions for lightning prediction. Experimental results show that the SCE-RCNN model achieved a critical success index (CSI) of 0.83, a probability of detection (POD) of 0.991, and a false alarm rate (FAR) reduced to 0.351, outperforming conventional deep learning models across multiple prediction metrics. This research provides reliable lightning forecasts to support real-time decision-making, making significant contributions to aviation safety, outdoor event planning, and disaster risk management. The model’s high accuracy and low false alarm rate highlight its value in both academic research and practical applications.

MCIDN: Deblurring Network for Metal Corrosion Images

The analysis of corrosion images is crucial in materials science, where acquiring clear images is fundamental for subsequent analysis. The goal of deblurring metal corrosion images is to reconstruct clear images from degraded ones. To the best of our knowledge, this study introduces the first paired blurry-sharp image dataset specifically designed for the metal corrosion domain, filling a critical gap in the existing research. This innovative approach effectively addresses the unique challenges associated with deblurring metal corrosion images. We propose a novel metal corrosion images deblurring network (MCIDN) that employs a dual-domain attention mechanism, integrating both spatial and frequency domains to enhance image clarity. This innovative approach effectively addresses the unique challenges associated with deblurring metal corrosion images. While self-attention is widely used in visual tasks, its quadratic complexity often leads to high computational costs. To address this issue, we introduce a new spatial channel attention module (SCAM) that employs dynamic group convolutions to achieve self-attention, effectively integrating information from local regions and enhancing representation learning capabilities. Recognizing the critical role of frequency components in image restoration, we develop a frequency channel attention module (FCAM) that selectively focuses on different frequency components of images, thereby enhancing deblurring performance. These two attention modules are seamlessly integrated into our network. Compared to existing methods, our approach demonstrates superior performance on datasets of blurry metal corrosion images, achieving a peak signal-to-noise ratio (PSNR) of 32.8645 dB and a structural similarity (SSIM) of 0.9768. These metrics indicate that our method provides clearer and more detailed reconstructions, significantly enhancing the image quality.

Ground-Based Cloud Image Segmentation Method Based on Improved U-Net

Cloud image segmentation is a technique that divides images captured by meteorological satellites or ground-based observations into different regions or categories. By extracting the distribution, shape, and dynamic features of clouds, it provides precise data support for the meteorological and environmental fields, significantly influencing photovoltaic (PV) power generation forecasting, astronomical telescope observatory site selection, and weather forecasting. A ground-based cloud image segmentation model based on an improved U-Net is proposed, which adopts an overall encoder–decoder structure. In the encoder phase, this paper constructs a dilated convolution–atrous spatial pyramid pooling (ASPP)–dilated convolution structure to enhance early cloud feature extraction. Dilated convolution is a novel type of convolution that expands the receptive field by inserting holes into standard convolution, thereby capturing a larger range of contextual information. ASPP maintains high resolution while paying attention to both local details and global structures of the image. In the decoder stage, the bicubic interpolation method is used for up-sampling to restore the feature map resolution and improve the clarity of the segmented image. The bicubic interpolation method refers to the use of cubic polynomial functions to interpolate the pixel values of the input image. In addition, this paper designs a novel skip connection layer structure between the encoder and decoder, composed of a depthwise separable path (DS path) and an improved channel spatial attention module (Im-CSAM) connected in sequence. The DS path combines depthwise separable convolutions and residual structures to facilitate information exchange between high-level and low-level features. The Im-CSAM is a modular attention mechanism that focuses on important cloud features in spatial and channel dimensions to enhance segmentation accuracy. Experiments show that compared to the traditional U-Net, the accuracy, precision, and MIoU of this model improved by 2.2%, 4.1%, and 5.0%, respectively, in the SWINySEG dataset, and by 3.2%, 3.6%, and 5.8%, respectively, in the TCDD dataset, proving that the improved method has a better generalization ability and segmentation performance.

Innovative Ghost Channel Spatial Attention Network with Adaptive Activation for Efficient Rice Disease Identification

To address the computational complexity and deployment challenges of traditional convolutional neural networks in rice disease identification, this paper proposes an efficient and lightweight model: Ghost Channel Spatial Attention ShuffleNet with Mish-ReLU Adaptive Activation Function (GCA-MiRaNet). Based on ShuffleNet V2, we effectively reduced the model’s parameter count by streamlining convolutional layers, decreasing stacking depth, and optimizing output channels. Additionally, the model incorporates the Ghost Module as a replacement for traditional 1 × 1 convolutions, further reducing computational overhead. Innovatively, we introduce a Channel Spatial Attention Mechanism (CSAM) that significantly enhances feature extraction and generalization aimed at rice disease detection. Through combining the advantages of Mish and ReLU, we designed the Mish-ReLU Adaptive Activation Function (MAAF), enhancing the model’s generalization capacity and convergence speed. Through transfer learning and ElasticNet regularization, the model’s accuracy has notably improved while effectively avoiding overfitting. Sufficient experimental results indicate that GCA-MiRaNet attains a precision of 94.76% on the rice disease dataset, with a 95.38% reduction in model parameters and a compact size of only 0.4 MB. Compared to traditional models such as ResNet50 and EfficientNet V2, GCA-MiRaNet demonstrates significant advantages in overall performance, especially on embedded devices. This model not only enables efficient and accurate real-time disease monitoring but also provides a viable solution for rice field protection drones and Internet of Things management systems, advancing the process of contemporary agricultural smart management.

A lightweight adaptive spatial channel attention efficient net B3 based generative adversarial network approach for MR image reconstruction from under sampled data.

Magnetic Resonance Imaging (MRI) stands out as a notable non-invasive method for medical imaging assessments, widely employed in early medical diagnoses due to its exceptional resolution in portraying soft tissue structures. However, the MRI method faces challenges with its inherently slow acquisition process, stemming from the sequential sampling in k-space and limitations in traversal speed due to physiological and hardware constraints. Compressed Sensing in MRI (CS-MRI) accelerates image acquisition by utilizing greatly under-sampled k-space information. Despite its advantages, conventional CS-MRI encounters issues such as sluggish iterations and artefacts at higher acceleration factors. Recent advancements integrate deep learning models into CS-MRI, inspired by successes in various computer vision domains. It has drawn significant attention from the MRI community because of its great potential for image reconstruction from undersampled k-space data in fast MRI. This paper proposes a lightweight Adaptive Spatial-Channel Attention EfficientNet B3-based Generative Adversarial Network (ASCA-EffNet GAN) for fast, high-quality MR image reconstruction from greatly under-sampled k-space information in CS-MRI. The proposed GAN employs a U-net generator with ASCA-based EfficientNet B3 for encoder blocks and a ResNet decoder. The discriminator is a binary classifier with ASCA-based EfficientNet B3, a fully connected layer and a sigmoid layer. The EfficientNet B3 utilizes a compound scaling strategy that achieves a balance amongst model depth, width, and resolution, resulting in optimal performance with a reduced number of parameters. Furthermore, the adaptive attention mechanisms in the proposed ASCA-EffNet GAN effectively capture spatial and channel-wise features, contributing to detailed anatomical structure reconstruction. Experimental evaluations on the dataset demonstrate ASCA-EffNet GAN's superior performance across various metrics, surpassing conventional reconstruction methods. Hence, ASCA-EffNet GAN showcases remarkable reconstruction capabilities even under high under-sampling rates, making it suitable for clinical applications.

CM-SC: Cross-modal spatial-channel attention network for image captioning

CM-SC: Cross-modal spatial-channel attention network for image captioning

Semi-supervised anomaly traffic detection via multi-frequency reconstruction

Anomaly traffic detection is a crucial issue in cyber-security field. Traditionally, many researchers have approached anomaly traffic detection as a supervised classification problem. However, in real-world scenarios, anomaly network traffic is unpredictable, dynamic, and difficult to collect. To address these challenges, we adopt an anomaly detection setting that trains using only normal traffic data and propose a novel semi-supervised method: Multi-Frequency Reconstruction (MFR). Our approach is driven by a key observation: traffic images often lack explicit patterns and texture features, making accurate modeling of normal traffic difficult. To overcome this limitation, we employ low-pass filters to extract multi-scale low-frequency information, offering a more effective multi-view perspective to identify anomaly patterns. Additionally, we introduce a channel-spatial attention mechanism within reconstruction models to enhance the network’s ability to capture the spatio-temporal features of traffic flows. Ultimately, to adequately leverage multi-view information, we integrate anomaly scores from multi-frequency branches, achieving a more comprehensive anomaly detection. Extensive experiments on three public anomaly traffic detection datasets demonstrate the superior performance of our method, outperforming state-of-the-art approaches by 4.6% and 10.5% in AUROC on the DataCon2020 and CIC-IDS2017 datasets, respectively.

LCANet: a model for analysis of students real-time sentiment by integrating attention mechanism and joint loss function

By recognizing students’ facial expressions in actual classroom situations, the students’ emotional states can be quickly uncovered, which can help teachers grasp the students’ learning rate, which allows teachers to adjust their teaching strategies and methods, thus improving the quality and effectiveness of classroom teaching. However, most previous facial expression recognition methods have problems such as missing key facial features and imbalanced class distributions in the dateset, resulting in low recognition accuracy. To address these challenges, this paper proposes LCANet, a model founded on a fused attention mechanism and a joint loss function, which allows the recognition of students’ emotions in real classroom scenarios. The model uses ConvNeXt V2 as the backbone network to optimize the global feature extraction capability of the model, and at the same time, it enables the model to pay closer attention to the key regions in facial expressions. We incorporate an improved Channel Spatial Attention (CSA) module as a way to extract more local feature information. Furthermore, to mitigate the class distribution imbalance problem in the facial expression dataset, we introduce a joint loss function. The experimental results show that our LCANet model has good recognition rates on both the public emotion datasets FERPlus, RAF-DB and AffectNet, with accuracies of 91.43%, 90.03% and 64.43%, respectively, with good robustness and generalizability. Additionally, we conducted experiments using the model in real classroom scenarios, detecting and accurately predicting students’ classroom emotions in real time, which provides an important reference for improving teaching in smart teaching scenarios.

Multi-Head Attention Affinity Diversity Sharing Network for Facial Expression Recognition

Facial expressions exhibit inherent similarities, variability, and complexity. In real-world scenarios, challenges such as partial occlusions, illumination changes, and individual differences further complicate the task of facial expression recognition (FER). To further improve the accuracy of FER, a Multi-head Attention Affinity and Diversity Sharing Network (MAADS) is proposed in this paper. MAADS comprises a Feature Discrimination Network (FDN), an Attention Distraction Network (ADN), and a Shared Fusion Network (SFN). To be specific, FDN first integrates attention weights into the objective function to capture the most discriminative features by using the proposed sparse affinity loss. Then, ADN employs multiple parallel attention networks to maximize diversity within spatial attention units and channel attention units, which guides the network to focus on distinct, non-overlapping facial regions. Finally, SFN deconstructs facial features into generic parts and unique parts, which allows the network to learn the distinctions between these features without having to relearn complete features from scratch. To validate the effectiveness of the proposed method, extensive experiments were conducted on several widely used in-the-wild datasets including RAF-DB, AffectNet-7, AffectNet-8, FERPlus, and SFEW. MAADS achieves the accuracy of 92.93%, 67.14%, 64.55%, 91.58%, and 62.41% on these datasets, respectively. The experimental results indicate that MAADS not only outperforms current state-of-the-art methods in recognition accuracy but also has a relatively low computational complexity.

Recognition of palaeo-glacial troughs with cirques on parts of Tibetan Plateau using multi-modal datasets with deep learning models

Recognition of landforms associated with past glaciation is crucial for understanding past ice dynamics and their relationship to climate. With the development of artificial intelligence technology, deep learning techniques have assisted in the automatic extraction of glacial landforms, but these methods still face problems of low precision and weak transferability. This study proposes a new method named geomorphology-attention DeeplabV3+ (GA-DeeplabV3+) model. This method adds spatial attention and channel attention modules based on the DeeplabV3+ network and utilizes a combination of multi-modal geographic data. Verification shows that the model proposed in this paper significantly enhances the recognition precision of glacial troughs with cirques compared to existing models, reaching a Mean Intersection over Union (MIoU) of 86.2% and a mean Pixel Accuracy (mPA) of 90.64% in the area of the Palaeo-Daocheng ice cap. In addition, validation experiments were conducted in the Peiku Gangri region and the Tenasserim mountains, achieving MIoU scores of 70.09% and 73.28% respectively. This accomplishment represents a vital stride towards automating the extraction of palaeo-glacial landforms, which holds great significance for analyzing the scale and evolution of ancient glaciers.

Diffusion network with spatial channel attention infusion and frequency spatial attention for brain tumor segmentation.

Accurate segmentation of gliomas is crucial for diagnosis, treatment planning, and evaluating therapeutic efficacy. Physicians typically analyze and delineate tumor regions in brain magnetic resonance imaging (MRI) images based on personal experience, which is often time-consuming and subject to individual interpretation. Despite advancements in deep learning technology for image segmentation, current techniques still face challenges in clearly defining tumor boundary contours and enhancing segmentationaccuracy. To address these issues, this paper proposes a conditional diffusion network (SF-Diff) with a spatial channel attention infusion (SCAI) module and a frequency spatial attention (FSA) mechanism to achieve accurate segmentation of the whole tumor (WT) region in braintumors. SF-Diff initially extracts multiscale information from multimodal MRI images and subsequently employs a diffusion model to restore boundaries and details, thereby enabling accurate brain tumor segmentation (BraTS). Specifically, a SCAI module is developed to capture multiscale information within and between encoder layers. A dual-channel upsampling block (DUB) is designed to assist in detail recovery during upsampling. A FSA mechanism is introduced to better match the conditional features with the diffusion probability distribution information. Furthermore, a cross-model loss function has been implemented to supervise the feature extraction of the conditional model and the noise distribution of the diffusionmodel. The dataset used in this paper is publicly available and includes 369 patient cases from the Multimodal BraTS Challenge 2020 (BraTS2020). The conducted experiments on BraTS2020 demonstrate that SF-Diff performs better than other state-of-the-art models. The method achieved a Dice score of 91.87%, a Hausdorff 95 of 5.47 mm, an IoU of 84.96%, a sensitivity of 92.29%, and a specificity of 99.95% onBraTS2020. The proposed SF-Diff performs well in identifying the WT region of the brain tumors compared to other state-of-the-art models, especially in terms of boundary contours and non-contiguous lesion regions, which is clinically significant. In the future, we will further develop this method for brain tumor three-class segmentation task.

Evaluation of a Task-Specific Self-Supervised Learning Framework in Digital Pathology Relative to Transfer Learning Approaches and Existing Foundation Models

An integral stage in typical digital pathology workflows involves deriving specific features from tiles extracted from a tessellated whole-slide image. Notably, various computer vision neural network architectures, particularly the ImageNet pretrained, have been extensively used in this domain. This study critically analyzes multiple strategies for encoding tiles to understand the extent of transfer learning and identify the most effective approach. The study categorizes neural network performance into 3 weight initialization methods: random, ImageNet-based, and self-supervised learning. Additionally, we propose a framework based on task-specific self-supervised learning, which introduces a shallow feature extraction method, employing a spatial-channel attention block to glean distinctive features optimized for histopathology intricacies. Across 2 different downstream classification tasks (patch classification and weakly supervised whole-slide image classification) with diverse classification data sets, including colorectal cancer histology, Patch Camelyon, prostate cancer detection, The Cancer Genome Atlas, and CIFAR-10, our task-specific self-supervised encoding approach consistently outperforms other convolutional neural network–based encoders. The better performances highlight the potential of task-specific attention-based self-supervised training in tailoring feature extraction for histopathology, indicating a shift from using pretrained models originating outside the histopathology domain. Our study supports the idea that task-specific self-supervised learning allows domain-specific feature extraction, encouraging a more focused analysis.

Multi-scale attention in attention neural network for single image deblurring

Image deblurring, which eliminates blurring artifacts to recover details from a given input image, represents an important task for the computer vision field. Recently, the attention mechanism with deep neural networks (DNN) demonstrates promising performance of image deblurring. However, they have difficulty learning complex blurry and sharp relationships through a balance of spatial detail and high-level contextualized information. Moreover, most existing attention-based DNN methods fail to selectively exploit the information from attention and non-attention branches. To address these challenges, we propose a new approach called Multi-Scale Attention in Attention (MSAiA) for image deblurring. MSAiA incorporates dynamic weight generation by leveraging the joint dependencies of channel and spatial information, allowing for adaptive changes to the weight values in attention and non-attention branches. In contrast to existing attention mechanisms that primarily consider channel or spatial dependencies and do not adequately utilize the information from attention and non-attention branches, our proposed AiA design combines channel-spatial attention. This attention mechanism effectively utilizes the dependencies between channel-spatial information to allocate weight values for attention and non-attention branches, enabling the full utilization of information from both branches. Consequently, the attention branch can more effectively incorporate useful information, while the non-attention branch avoids less useful information. Additionally, we employ a novel multi-scale neural network that aims to learn the relationships between blurring artifacts and the original sharp image by further exploiting multi-scale information. The experimental results prove that the proposed MSAiA achieves superior deblurring performance compared with the state-of-the-art methods.

ACU-TransNet: Attention and convolution-augmented UNet-transformer network for polyp segmentation.

UNet has achieved great success in medical image segmentation. However, due to the inherent locality of convolution operations, UNet is deficient in capturing global features and long-range dependencies of polyps, resulting in less accurate polyp recognition for complex morphologies and backgrounds. Transformers, with their sequential operations, are better at perceiving global features but lack low-level details, leading to limited localization ability. If the advantages of both architectures can be effectively combined, the accuracy of polyp segmentation can be further improved. In this paper, we propose an attention and convolution-augmented UNet-Transformer Network (ACU-TransNet) for polyp segmentation. This network is composed of the comprehensive attention UNet and the Transformer head, sequentially connected by the bridge layer. On the one hand, the comprehensive attention UNet enhances specific feature extraction through deformable convolution and channel attention in the first layer of the encoder and achieves more accurate shape extraction through spatial attention and channel attention in the decoder. On the other hand, the Transformer head supplements fine-grained information through convolutional attention and acquires hierarchical global characteristics from the feature maps. mcU-TransNet could comprehensively learn dataset features and enhance colonoscopy interpretability for polyp detection. Experimental results on the CVC-ClinicDB and Kvasir-SEG datasets demonstrate that mcU-TransNet outperforms existing state-of-the-art methods, showcasing its robustness.

HDKD: Hybrid data-efficient knowledge distillation network for medical image classification

Vision Transformers (ViTs) have achieved significant advancement in computer vision tasks due to their powerful modeling capacity. However, their performance notably degrades when trained with insufficient data due to lack of inherent inductive biases. Distilling knowledge and inductive biases from a Convolutional Neural Network (CNN) teacher has emerged as an effective strategy for enhancing the generalization of ViTs on limited datasets. Previous approaches to Knowledge Distillation (KD) have pursued two primary paths: some focused solely on distilling the logit distribution from CNN teacher to ViT student, neglecting the rich semantic information present in intermediate features due to the structural differences between them. Others integrated feature distillation along with logit distillation, yet this introduced alignment operations that limits the amount of knowledge transferred due to mismatched architectures and increased the computational overhead. To this end, this paper presents Hybrid Data-efficient Knowledge Distillation (HDKD) paradigm which employs a CNN teacher and a hybrid student. The choice of hybrid student serves two main aspects. First, it leverages the strengths of both convolutions and transformers while sharing the convolutional structure with the teacher model. Second, this shared structure enables the direct application of feature distillation without any information loss or additional computational overhead. Additionally, we propose an efficient light-weight convolutional block named Mobile Channel-Spatial Attention (MBCSA), which serves as the primary convolutional block in both teacher and student models. Extensive experiments on two medical public datasets showcase the superiority of HDKD over other state-of-the-art models and its computational efficiency. Source code at: https://github.com/omarsherif200/HDKD.

BIF-Net: Boundary information fusion network for abdominal aortic aneurysm segmentation

The accurate abdominal aortic aneurysm (AAA) segmentation is significant for assisting clinicians in diagnosis and treatment planning. However, existing segmentation methods exhibit a low utilization rate for the semantic information of vessel boundaries, which is disadvantageous for segmenting AAA with significant scale variability of vessel diameter (diameter ranges from 4 mm to 85 mm). To tackle this problem, we introduce a boundary information fusion network (BIF-Net) specially designed for AAA segmentation. BIF-Net initially constructs convolutional kernel functions based on Gabor and Sobel operators, enriching the global semantic features and localization information through the Gabor and Sobel dilated convolution (GSDC) module. Additionally, BIF-Net supplements lost boundary feature information during the sampling process through the guided filtering feature supplementation (GFFS) module and the channel-spatial attention module (CSAM), enhancing the ability to capture targets with shape diversity and boundary features. Finally, we introduce a boundary feature loss function to alleviate the impact of the imbalance between positive and negative samples. The results demonstrate that BIF-Net outperforms current state-of-the-art methods across multiple evaluation metrics, achieving the highest Dice similarity coefficient (DSC) accuracies of 93.29 % and 91.01 % on the preoperative and postoperative datasets, respectively. Compared to the state-of-the-art methods, BIF-Net improves DSC by 6.86 % and 3.85 %. Due to the powerful boundary feature extraction ability, the proposed BIF-Net is a competitive AAA segmentation method exhibiting significant potential for application in diagnosis and treatment processes of AAA.