A Comparison of Edge Constrained Optical Reconstruction Methods Incorporating Spectral and MR- Derived Spatial Information

摘要

Incorporating near infrared (NIR) diffuse optical tomography into magnetic resonance imaging (MRI) increases the value of MR breast cancer imaging because it adds functional imaging of hemoglobin, oxygen saturation, water, lipid content, and scattering parameters, properties that infer tissue health. Reconstruction algorithms that incorporate MR into a diffusive modality accrue unavoidable errors from improper tissue segmentation of the MR image, which create inaccuracies in the structural prior. This paper focuses on identifying the most accurate reconstruction approach based on imperfect prior knowledge of tissue boundaries. Specifically, it focuses on how unavoidable segmentation errors of different breast densities affect edge-constraining reconstruction methods to determine the correct approach. Results show that these reconstruction methods all retain the improperly defined edges, but are quantitatively accurate even when the anatomical boundaries mismatch the optical boundaries by as much as 50%. The most accurate approach is one where the problem has been reduced to the least number of unknowns, and the edges are constrained through regularization.