Dimension Reduction Based on Grouped Feature Selection Strategy for Fluorescence Molecular Tomography

Fluorescence Molecular Tomography (FMT) is an optical molecular imaging technique with high sensitivity. Multi-point excitation and multi-view measurement have improved the recovered results while also resulting in high computation and memory requirements. In this work, we proposed a grouped feature selection strategy (GFS) based on correlation and information entropy to reduce the size of the system matrix. It further reduced the computational complexity. Firstly, a similarity is constructed based on the correlation between the matrix rows. Then a group is created by the surface measurements. Secondly, information entropy is defined to measure the information contained in the surface measurement. Finally, these two measurement methods are combined with minimum redundancy and maximum relevance (mRMR) to construct the optimal subset of the original data, which completes the dimensionality reduction of the matrix. Numerical simulation experiments show that GFS outperforms the traditional principal component analysis (PCA) method in terms of reconstruction accuracy and reconstruction efficiency.Clinical Relevance— The primary objective of this work is to enhance the reconstruction speed of FMT. Current experiments are focused on simulation experiments, while the proposed downscaling strategy will facilitate rapid clinical diagnosis and treatment decisions in the future