文献阅读3

05

Date:2022.12.17--05
Title:Multi‑Modal Feature Fusion‑Based Multi‑Branch Classification Network for Pulmonary Nodule Malignancy Suspiciousness Diagnosis Link:Multi-Modal Feature Fusion-Based Multi-Branch Classification Network for Pulmonary Nodule Malignancy Suspiciousness Diagnosis | SpringerLink ​
Framework
Methodology	Multi-branch:Three branches with different receptive fields at different levels (shallow layer, middle layer, and deep layer) are adopted to carry out multi-scale fusion of feature and improve the accuracy of classification. 3D ECA-ResNet:Two fully connected layers are replaced by fast one-dimensional convolution with kernel size k .
Results
Contributions	1.Radiological data of pulmonary nodules are used to construct a structured modal feature vector. 2.Multi-branch fusion network combined with 3D ECAResNet is used to extract unstructured modal features from CT images. 3.Multi-modal feature fusion of structured data and unstructured data is performed to distinguish benign and malignant nodules.
Conclusion	The multi-branch fusion enhances the multiscale feature extraction ability of pulmonary nodules, and 3D ECA-ResNet dynamically adjusts channel features to further extract more representative nodule features.
Notes	1.Three branches with different receptive fields at different levels (shallow layer, middle layer, and deep layer. 2.Two fully connected layers are replaced by fast one-dimensional convolution with kernel size k .