Convolutional Neural Networks(CNNs)have shown remarkable capabilities in extracting local features from images,yet they often overlook the underlying relationships between pixels.To address this limitation,previous ap...Convolutional Neural Networks(CNNs)have shown remarkable capabilities in extracting local features from images,yet they often overlook the underlying relationships between pixels.To address this limitation,previous approaches have attempted to combine CNNs with Graph Convolutional Networks(GCNs)to capture global features.However,these approaches typically neglect the topological structure information of the graph during the global feature extraction stage.This paper proposes a novel end-to-end hybrid architecture called the Multi-Graph Pooling Network(MGPN),which is designed explicitly for chest X-ray image classification.Our approach sequentially combines CNNs and GCNs,enabling the learning of both local and global features from individual images.Recognizing that different nodes contribute differently to the final graph representation,we introduce an NI-GTP module to enhance the extraction of ultimate global features.Additionally,we introduce a G-LFF module to fuse the local and global features effectively.展开更多
The pooling operation is used in graph classification tasks to leverage hierarchical structures preserved in data and reduce computational complexity.However,pooling shrinkage discards graph details,and existing pooli...The pooling operation is used in graph classification tasks to leverage hierarchical structures preserved in data and reduce computational complexity.However,pooling shrinkage discards graph details,and existing pooling methods may lead to the loss of key classification features.In this work,we propose a residual convolutional graph neural network to tackle the problem of key classification features losing.Particularly,our contributions are threefold:(1)Different from existing methods,we propose a new strategy to calculate sorting values and verify their importance for graph classification.Our strategy does not only use features of simple nodes but also their neighbors for the accurate evaluation of its importance.(2)We design a new graph convolutional layer architecture with the residual connection.By feeding discarded features back into the network architecture,we reduce the probability of losing critical features for graph classification.(3)We propose a new method for graph-level representation.The messages for each node are aggregated separately,and then different attention levels are assigned to each node and merged into a graph-level representation to retain structural and critical information for classification.Our experimental results show that our method leads to state-of-the-art results on multiple graph classification benchmarks.展开更多
基金supported by the National Natural Science Foundation of China(62276092,62303167)Key Science and Technology Program of Henan Province(212102310084)+11 种基金MRC(MC_PC_17171)Royal Society(RP202G0230)BHF(AA/18/3/34220)Hope Foundation for Cancer Research(RM60G0680)GCRF(P202PF11).Sino-UK Industrial Fund(RP202G0289)LIAS(P202ED10,P202RE969)Key Scientific Research Projects of Colleges and Universities in Henan Province(25A520009)Data Science Enhancement Fund(P202RE237)Fight for Sight(24NN201)Sino-UK Education Fund(OP202006)BBSRC(RM32G0178B8).
文摘Convolutional Neural Networks(CNNs)have shown remarkable capabilities in extracting local features from images,yet they often overlook the underlying relationships between pixels.To address this limitation,previous approaches have attempted to combine CNNs with Graph Convolutional Networks(GCNs)to capture global features.However,these approaches typically neglect the topological structure information of the graph during the global feature extraction stage.This paper proposes a novel end-to-end hybrid architecture called the Multi-Graph Pooling Network(MGPN),which is designed explicitly for chest X-ray image classification.Our approach sequentially combines CNNs and GCNs,enabling the learning of both local and global features from individual images.Recognizing that different nodes contribute differently to the final graph representation,we introduce an NI-GTP module to enhance the extraction of ultimate global features.Additionally,we introduce a G-LFF module to fuse the local and global features effectively.
基金supported by the National Natural Science Foundation of China(No.62072335)the Tianjin Science and Technology Program(No.19PTZWHZ00020)。
文摘The pooling operation is used in graph classification tasks to leverage hierarchical structures preserved in data and reduce computational complexity.However,pooling shrinkage discards graph details,and existing pooling methods may lead to the loss of key classification features.In this work,we propose a residual convolutional graph neural network to tackle the problem of key classification features losing.Particularly,our contributions are threefold:(1)Different from existing methods,we propose a new strategy to calculate sorting values and verify their importance for graph classification.Our strategy does not only use features of simple nodes but also their neighbors for the accurate evaluation of its importance.(2)We design a new graph convolutional layer architecture with the residual connection.By feeding discarded features back into the network architecture,we reduce the probability of losing critical features for graph classification.(3)We propose a new method for graph-level representation.The messages for each node are aggregated separately,and then different attention levels are assigned to each node and merged into a graph-level representation to retain structural and critical information for classification.Our experimental results show that our method leads to state-of-the-art results on multiple graph classification benchmarks.
