Seed based correlation analysis is highly depend on where the seeds sit, also highly depend on the spatial normalization algorithm. If the subjects are not coregistered with each other well, then seed based correlation analysis can be biased.
Here I proposed a quick idea to achieve functional connectivity analysis has a characteristic of seed based correlation analysis, but also take some idea from ICA (especially dual regression for ICA). I haven't checked literature for this quick idea yet, nor implemented it (I wrote half of the program in July, 2014, and hope I can release it soon).
It's essentially to create a subject specific network integration map, together with a time course of how each time point contribute to that network.
1. Do a seed based correlation analysis, for example, choose PCC seed for default mode network analysis. This step resulted a functional connectivity map. Alternatively, could be started with Step 2 if using an existing network template (e.g., use network template from Smith et al., 2009, PNAS).
2. For each time point, calculate it's spatial correlation with the FC map acquired in Step 1 or a network template, resulting in a time course. This time course indicate the contribution (involvement, or integration) of that time point to that network.
3. Use the time course acquired in Step 2 as a seed course, re-calculate a FC map.
4. Repeat Steps 2 and 3, until convergent, e.g., the correlation between the FC map at the current iteration and FC map at the previous iteration is greater than a setting number, e.g., r>0.99999.
5. The final FC map means a subject-specific network, the value at each voxel means how that voxel contributed (involved) in that network. The final time course means how the network fluctuates across time.
I hope this strategy can be more robust than simple seed based correlation analysis. Also hope some researchers interested in this idea can follow up, and even conduct a study.
Thanks,
Chao-Gan