Multi-view Information Fusion Methods in CEM and Mammographic Images: A Comprehensive Review

The combination of multi-view information is crucial in both CEM images and mammographic images. Currently, there is no systematic and comprehensive method for combining multi-view information in CEM images. In mammographic images, there are many methods for combining multi-view information, but most of them use a simple strategy of feature concatenation. For example, Sun et al. use two independent CNN networks to extract features from CC and MLO views, and then concatenate the features and input them to the subsequent network. Nasir et al. use four independent CNN networks to extract features from four views, and then concatenate the features after the global pooling layer, followed by a fully connected (fc) layer to obtain the final result. Wu et al. also concatenate the features after global pooling, but their work has two differences compared to Nasir et al.: (1) they use two consecutive fc layers after feature concatenation to better learn the relationship between different view features; (2) all CC views share one feature extraction network, and all MLO views share another feature extraction network. In addition, they also compare the performance of different fusion strategies. Li et al. compare the performance of extracting features from different views using independent model parameters and shared model parameters, and the conclusion is that the network can achieve better results when using shared model parameters to extract features from different views. Other networks also use feature concatenation strategy at different positions in the model to fuse multi-view information (e.g., Liu et al., Yan et al., Yang et al.). Due to the potential misalignment of different views, most concatenation-based methods fuse multi-view information after the global pooling layer, resulting in the loss of a large amount of local information and the inability to fully utilize multi-view information.