Learning a new method, such as regional homogeneity analysis, can be quite difficult, and one often asks whether there is an easier, quicker method to become enlightened. Unfortunately, such learning can only be accomplished through large, dense books. Specifically, you should go to the library, check out the largest, heaviest book on regional homogeneity analysis you can find, and then go to the lab of someone smarter than you and threaten to smash their computer with the book unless they do the analysis for you.
If for some reason that isn't an option, the next best way is to read how others have implemented the same analysis; such as me, for example. Just because I haven't published anything on this method, and just because I am learning it for the first time, doesn't mean you should go do something rash, such as try to figure it out on your own. Rather, come along as we attempt to unravel the intriguing mystery of regional homogeneity analysis, and hide from irate postdocs whose computers we have destroyed. In addition to the thrills and danger of finding things out, if you follow all of the steps outlined in this multi-part series, I promise that you will be the first one to learn this technique from a blog. And surely, that must count for something.
With regional homogeneity analysis (or ReHo), researchers ask similar questions as with functional connectivity analysis; however, in the case of ReHo, we correlate the timecourse in one voxel with its immediate neighbors, or with a range of neighbors within a specified radius, instead of using a single voxel or seed region and testing for correlations with every other voxel in the brain, as in standard functional connectivity analysis.
As an analogy, think of ReHo as searching for similarities in the timecourse of the day's temperature between different counties across a country. One area's temperature timecourse will be highly correlated with neighboring counties's temperatures, and the similarity will tend to decrease the further away you go from the county you started in. Functional connectivity analysis, on the other hand, looks at any other county that shows a similar temperature timecourse to the county you are currently in.
Similarly, when ReHo is applied to functional data, we look for differences in local connectivity; that is, whether there are differences in connectivity within small areas or cortical regions. For example, when comparing patient groups to control groups, there may be significantly less or significantly more functional connectivity in anterior and posterior cingulate areas, possibly pointing towards some deficiency or overexcitation of communication within those areas. (Note that any differences found in any brain area with the patient group implies that there is obviously something "wrong" with that particular area compared to the control group, and that the opposite can never be true. While I stand behind this arbitrary judgment one hundred percent, I would also appreciate it if you never quoted me on this.)
As with the preprocessing step of smoothing, ReHo is applied to all voxels simultaneously, and that the corresponding correlation statistic in each voxel quantifies how much it correlates with its neighboring voxels. This correlation statistic is called Kendall's W, and ranges between 0 (no correlation at all between the specified voxel and its neighbors) and 1 (perfect correlation with all neighbors). Once these maps are generated, they can then be normalized and entered into t-tests, producing similar maps that we used with our functional connectivity analysis.
Now that we have covered this technique in outline, in our next post we will move on to the second, more difficult part: Kidnapping a senior research assistant and forcing him to do the analysis for us.
No, wait! What I meant was, we will review some papers that have used ReHo, and attempt to apply the same steps to our own analysis. If you have already downloaded and processed the KKI data that we used for our previous tutorial on functional connectivity, we will be applying a slightly different variation to create our ReHo analysis stream - one which will, I hope, not include federal crimes or destroying property.
If for some reason that isn't an option, the next best way is to read how others have implemented the same analysis; such as me, for example. Just because I haven't published anything on this method, and just because I am learning it for the first time, doesn't mean you should go do something rash, such as try to figure it out on your own. Rather, come along as we attempt to unravel the intriguing mystery of regional homogeneity analysis, and hide from irate postdocs whose computers we have destroyed. In addition to the thrills and danger of finding things out, if you follow all of the steps outlined in this multi-part series, I promise that you will be the first one to learn this technique from a blog. And surely, that must count for something.
With regional homogeneity analysis (or ReHo), researchers ask similar questions as with functional connectivity analysis; however, in the case of ReHo, we correlate the timecourse in one voxel with its immediate neighbors, or with a range of neighbors within a specified radius, instead of using a single voxel or seed region and testing for correlations with every other voxel in the brain, as in standard functional connectivity analysis.
As an analogy, think of ReHo as searching for similarities in the timecourse of the day's temperature between different counties across a country. One area's temperature timecourse will be highly correlated with neighboring counties's temperatures, and the similarity will tend to decrease the further away you go from the county you started in. Functional connectivity analysis, on the other hand, looks at any other county that shows a similar temperature timecourse to the county you are currently in.
Similarly, when ReHo is applied to functional data, we look for differences in local connectivity; that is, whether there are differences in connectivity within small areas or cortical regions. For example, when comparing patient groups to control groups, there may be significantly less or significantly more functional connectivity in anterior and posterior cingulate areas, possibly pointing towards some deficiency or overexcitation of communication within those areas. (Note that any differences found in any brain area with the patient group implies that there is obviously something "wrong" with that particular area compared to the control group, and that the opposite can never be true. While I stand behind this arbitrary judgment one hundred percent, I would also appreciate it if you never quoted me on this.)
As with the preprocessing step of smoothing, ReHo is applied to all voxels simultaneously, and that the corresponding correlation statistic in each voxel quantifies how much it correlates with its neighboring voxels. This correlation statistic is called Kendall's W, and ranges between 0 (no correlation at all between the specified voxel and its neighbors) and 1 (perfect correlation with all neighbors). Once these maps are generated, they can then be normalized and entered into t-tests, producing similar maps that we used with our functional connectivity analysis.
Now that we have covered this technique in outline, in our next post we will move on to the second, more difficult part: Kidnapping a senior research assistant and forcing him to do the analysis for us.
No, wait! What I meant was, we will review some papers that have used ReHo, and attempt to apply the same steps to our own analysis. If you have already downloaded and processed the KKI data that we used for our previous tutorial on functional connectivity, we will be applying a slightly different variation to create our ReHo analysis stream - one which will, I hope, not include federal crimes or destroying property.