Correlation Tensor Decomposition and Its Application in Spatial Imaging Data

Abstract

Multi-dimensional tensor data have gained increasing attention in the recent years, especially in biomedical imaging analyses. However, the most existing tensor models are only based on the mean information of imaging pixels. Motivated by multimodal optical imaging data in a breast cancer study, we develop a new tensor learning approach to use pixel-wise correlation information, which is represented through the higher order correlation tensor. We proposed a novel semi-symmetric correlation tensor decomposition method which effectively captures the informative spatial patterns of pixel-wise correlations to facilitate cancer diagnosis. We establish the theoretical properties for recovering structure and for classification consistency. In addition, we develop an efficient algorithm to achieve computational scalability. Our simulation studies and an application on breast cancer imaging data all indicate that the proposed method outperforms other competing methods in terms of pattern recognition and prediction accuracy.

Keywords:

• Dimension reduction
• Image processing
• Multidimensional data
• Spatial correlation
• Tensor decomposition

Supplementary Materials

The online supplement contains the generalized model with cross-modality correlations, technical proofs, additional numerical results and computational details for CNN.

Acknowledgments

The authors are grateful to reviewers, the associate editor and editor for their insightful comments and suggestions which have improved the article significantly. The authors also appreciate Biophotonics Imaging Laboratory of University of Illinois, Urbana-Champaign for providing the breast cancer image data.

Additional information

Funding

The work is supported by NSF (grants DMS 1952406 and DMS 1821198).