DGR-Net: Deep Groupwise Registration of Multispectral Images

Groupwise registration of multispectral images (MSI) is clinically essential to facilitate accurate information fusion across different modalities. However, the groupwise registration of multispectral images is a challenging task because multiple different imaging modalities makes it difficult to jointly optimize the deformation. In this work, we propose an unbiased deep groupwise registration framework, DGR-Net, which takes a complete consideration of the information aggregated by calculating the deformation of the sequence image. Our framwork guided by principal component analysis (PCA) image. Network optimization is accelerated by combining internal smoothing and external correlation of the deformation fields. Experimental results have shown that, the proposed method can achieve promising accuracy and efficiency for the challenging multi-modality groupwise registration task and also outperforms the state-of-the-art approaches.