Multi-modal 3-D Medical Image Fusion Based on Tensor Robust Principal Component Analysis

Abstract

Multimodal 3-D medical image fusion is important for the clinical diagnosis. The common fusion operation processes each slice of 3-D medical image individually, that may ignore the 3-D information. To extract the 3-D information of multi-modal medical images effectively, we propose a Tensor Robust Principal Component Analysis (TRPCA) based 3-D medical image fusion method. The low rank and sparse components of TRPCA can reveal the 3-D structures of medical image, specially the correlation and differences among the adjacent slices. In addition, we adopt the “3-D weighted local Laplacian energy” rule and the “max-absolute” rule for low-rank components fusion and sparse components fusion, respectively. The developed 3-D weighted local Laplacian energy fusion rule can determine the activity level of 3-D features and preserve 3-D structural information effectively. The experimental results on various medical images, acquired in different modalities, show the performance of proposed method by comparing with several popular methods.