Prostate Diffusion Weighted-Magnetic Resonance Image analysis using Multivariate Curve Resolution methods

Abstract

Multivariate Curve Resolution (MCR) has been applied on prostate Diffusion Weighted

Sequential multiblock partial least squares discriminant analysis for assessing prostate cancer aggressiveness with multiparametric magnetic resonance imaging

“…A detailed explanation of the model definition and calculation can be found in Ref. [15]. Additionally, the RSS was also calculated as an additional biomarker for the same reasons (i.e.…”

“…D T is a matrix containing in its rows each of the pure behaviors (pure spectrum associated to each physiological phenomena); C gathers in its rows the relative contribution of each behavior for each pixel of the image; and E is a residual matrix [14][15][16]24].…”

“…To do so, it is mandatory to fit and characterize the intensity versus time (in the case of DCE-MR) or versus b-value (in the case of DW-MR) curves associated to each pixel of the images. Although pharmacokinetic models [8] (in the case of DCE-MR) and exponential models [9,10] (in the case of DW-MR) have the ability to provide clinically-oriented biomarkers in tumor analysis, their interpretation is not easy nor direct in many cases; so new biomarkers obtained from latent variables-based multivariate statistical models, like multivariate curve resolution (MCR [11][12][13]), have recently been also proposed [14][15][16][17] to improve prostate cancer assessment.…”

Application of multivariate image analysis to prostate cancer for improving the comprehension of the related physiological phenomena and the development and validation of new imaging biomarkers

“…This model does not impose these types of restrictions, but others related to some a priori knowledge, when available; hence being able to provide more clinically (or physiologically) interpretable results. Applications of MCR models on biomedical images can be found in [88][89][90][91].…”

“…Finally, once the biexponential model is validated, the shape constraints for the D matrix were included in the iterative process of the MCR-ALS algorithm, assuming classical exponential expressions ai exp(bi b) for both behaviors, and non-negativity for S and D matrices. Details can be found in [89]. For every case, the exponential parameters, ai and bi (i=1,2), obtained from Model 2 have been used as an initial approximation.…”