search

open-connectome-classes / StatConn-Spring-2015-Info

introductory material
18 stars 4 forks source link

In search of a mean connectome #131

Open indigorose1 opened 9 years ago

indigorose1 commented 9 years ago

When we were talking about finding a mean connectome at the end of class, there were a few options about what to make each node in the SBM from MRI data. We could make each node a voxel, but then each voxel doesn't necessarily line up with the same structure in each brain when compared to an atlas. But if we make each node a brain region, we lose resolution. My question is what would probably be the most useful. Is sacrificing resolution worth easier comparability, or can we still do something with the higher res crap?

dlee138 commented 9 years ago

I would say more information is better, because we can always process the high-res data to make low-res data that could be potentially useful for us. For example, we could cluster voxels to particular regions and use those regions as nodes instead of using voxels as nodes. This would give us easier comparability, and if something looks off or the results look insignificant, we can always go back to the high-res and try re-clustering the nodes again.

whock commented 9 years ago

I think it depends on what level of connectivity you're interested in studying and I agree that there's definitely a fundamental trade-off, as you seem to be suggesting. I think that having higher-res data and taking on the challenge is preferable to making do with lower-res data, though, because at least in the former case you have some task (i.e. registration with an atlas or template) that you can try to optimize. And even though the atlases, templates, etc differ from one another as long as you're consistent (pick one and stay with it) I think you can get at some interesting biology.

Another issue is what does the brain "think" the basic unit of connectivity is? For many nuclei and cortical areas, the cells within distinct areas of the nuclei/cortical areas are pretty similar in their connectivity. So maybe you don't even need the highest-res data possible.

yaxigeigei commented 9 years ago

It may not really matter that we couldn't align brains perfectly. There are intrinsic noises in every step of data acquisition and processing, including registration. This is why we use statistics.