Skip to main content
. Author manuscript; available in PMC: 2023 Sep 1.
Published in final edited form as: Biometrics. 2021 Nov 15;78(3):1113–1117. doi: 10.1111/biom.13591

Figure 1.

Figure 1.

Stability of DICA and ICA methods on fMRI (left) and DTI (right) data among 5 re-runs for L ICs. For fMRI, DICA with its original fixed-initialization ICA (DICA-FI) has better Dice similarity between runs than with random initialization (DICA-RI), but this is still better than standard ICA (ICA-RI). For DTI, there are two sets of results, one with L = 14 as recommended by the authors, and one with L = 6 to facilitate comparisons with standard ICA. DICA shows an expansive range of Dice similarities, from 0 to 1, with median Dice slightly better for L = 6; standard ICA is only slightly better. Overall, the Dice similarities are disappointing; while some ICs may be reliable, the near zero Dice values indicate that often ICs will be obtained on some runs that will simply not correspond to ICs found on other runs.