EAI Endorsed Transactions on e-Learning

Applications of Image Segmentation Techniques in Medical Images

Yang-yang Hou — 2024-07-19

Image segmentation is an important research direction in medical image processing tasks, and it is also a challenging task in the field of computer vision. At present, there have been many image segmentation methods, including traditional segmentation methods and deep learning-based segmentation methods. Through the understanding and learning of the current situation in the field of medical image segmentation, this paper systematically combs it. Firstly, it briefly introduces the traditional image segmentation methods such as threshold method, region method and graph cut method, and focuses on the commonly used network architectures based on deep learning such as CNN, FCN, U-Net, SegNet, PSPNet, Mask R-CNN. At the same time, the application in medical image segmentation is expounded. Finally, the challenges and development opportunities of medical image segmentation technology based on deep learning are discussed.

A Review of Hypergraph Neural Networks

Xinke Zhi — 2024-10-16

In recent years, Graph Neural Networks (GNNs) have seen notable success in fields such as recommendation systems and natural language processing, largely due to the availability of vast amounts of data and powerful computational resources. GNNs are primarily designed to work with graph data that involve pairwise relationships. However, in many real-world networks, the relationships between entities are complex and go beyond simple pairwise connections, as seen in scientific collaboration networks, protein networks, and similar domains. If these complex relationships are directly represented as pairwise relationships using graph structures, it can lead to information loss. A hypergraph, as a special kind of graph-structured data, can represent higher-order relationships that cannot be fully captured by graphs, thereby addressing the limitations of graphs. In light of this, researchers have begun to focus on how to design neural networks on hypergraphs, leading to the proposal of hypergraph neural network (HGNN) models for downstream tasks. Therefore, this paper reviews the existing hypergraph neural network models. The review is conducted from two perspectives: spectral analysis methods and neural network methods on hypergraphs, discussing both unfolded and non-unfolded methods, and further subdividing them based on their algorithm characteristics and application scenarios. Subsequently, the design concepts of various algorithms are analyzed and compared, and the advantages and disadvantages of each type of algorithm are summarized based on experimental results. Finally, potential future research directions in hypergraph learning are discussed.

A Community Detection Algorithm Based on Balanced Label Propagation

Huijuan Jia — 2024-07-16

OBJECTIVES: In conventional label propagation algorithms, the randomness inherent in the selection order of nodes and subsequent label propagation frequently leads to instability and reduces the accuracy of community detection outcomes.

METHODS: First, select the initial node according to the node importance and assign different labels to each initial node, aiming to reduce the number of iterations of the algorithm and improve the efficiency and stability of the algorithm; second, identify the neighbor node with the largest connection to each initial node for the pre-propagation of the labels; then, the algorithm traverses the nodes in descending order of the node importance for the propagation of labels to reduce the randomness of the label propagation process; finally, the final community is formed through the rapid merging of small communities.

RESULTS: The experimental results on multiple real datasets and artificially generated networks show that the stability and accuracy are all improved.

CONCLUSION: The proposed community detection algorithm based on balanced label propagation is better than the other four advanced algorithms on Q and NMI values of community division results.

ARM-Net: Improved MRI brain tumor segmentation method based on attentional mechanism and residual module

MingHu — 2024-07-26

INTRODUCTION: Accurate tumor segmentation is a prerequisite for reliable diagnosis and treatment of brain cancer. Gliomas, a highly prevalent and life-threatening type of brain tumor, pose a challenge for segmentation due to the intricate nature of brain structures and unpredictable appearances on brain MRI images.

OBJECTIVES: Current methods for brain tumor segmentation mostly rely on deep convolutional neural networks, which suffer from significant loss of feature information during encoding and decoding and the inability to capture tumor contours in detail.

METHODS: To address these challenges, this study rethinks the network architecture for MRI brain tumor segmentation. It proposes ARM-Net: an improved method for MRI brain tumor segmentation based on attention mechanisms and residual modules. Firstly, inverted external attention and dilated gated attention are employed in the last two layers of the encoder to enable the network to interact with both lesion areas and global information, facilitating better interaction among the four modalities. Secondly, different numbers of Res-Paths are added in the encoder's first two layers and the decoder's last two layers to effectively mitigate the semantic gap issues caused by traditional skip connections.

RESULTS: Experiments on the BraTS 2019 dataset demonstrate that ARM-Net outperforms other similar models in terms of segmentation performance.

CONCLUSION: The experiment showed that the ARM-Net model could segment the contour structure of the tumor better than other methods.