Fig. 2.
L-curve for ℓ2-regularized QSM results for a young subject. X-axis: data consistency term ||δ – F−1 DF χ||2 in regularized reconstruction for varying values of the smoothing parameter β. Y- axis: regularization term ||G χ||2. Setting β = 3·10−3 yielded an under-regularized susceptibility map with ringing artifacts (a), whereas using β = 7·10−2 resulted an over-regularized reconstruction (c). For β = 1.5·10−2, the operating point with the largest curvature on the L-curve was obtained (b). This setting was used for the reported ℓ2-regularized results.