Automated Principal Component Analysis (PCA) filtering for denoising DCE-MRI data

Introduction Dynamic Contrast Enhanced MRI (DCE-MRI) is recognized as an efficient diagnosis and prognostics tool in several lesions such as in solid cancers. However, the improvement of its quantification remains crucial. Among other questions, the signal to noise ratio (SNR) is known to be poor in DCE-MRI, yielding a low efficiency of the DCE-MRI to detect and characterize small o heterogeneous lesions. Conventional spatial or time low pass filters induce losses in spatial or time resolution providing possible quantification errors. Here an adaptative Principal Component Analysis (PCA) filtering process is proposed to avoid these limitations. The originality of the process consists in selecting automatically the number of PCA factors for a maximal denoising efficiency, which is acceptable for a minimal loss of physiological information.