help > seed-to-voxel coordinate for F-contrast
Showing 1-5 of 5 posts
Jan 7, 2017 06:01 PM | Theodore Turesky
seed-to-voxel coordinate for F-contrast
Hi Alfonso et al.,
Before describing my question, I want to thank you and the other CONN folks for providing this wonderful tool and for maintaining this forum. It's a remarkable feat!
Earlier today, I noticed a discrepancy in CONN's seed-to-voxel results explorer and I was hoping to receive some clarification.
I'm noticing that when I start with the effect of group (contrast vector: [1]) in the main GUI and then, in the results explorer, go from the one-sided positive (t) contrast to the two-sided (f) contrast, that the coordinates for the positive clusters stay the same. In contrast, if I compare the coordinates from the one-sided negative (t) contrast to the two-sided (f) contrast, that the coordinates for the negative clusters change.
If I type the F-contrast matrix into the main GUI (e.g., [1; -1] rather than [1]) or use the spm display button to define a new F-contrast, the coordinates are the same as both positive and negative one-sided contrasts. I'm wondering whether the positive one-sided coordinates should be updating when going from one- to two-sided or the negative one-sided coordinates are not updating accurately in the results explorer or I'm misinterpreting in some way. Any guidance you can provide would be greatly appreciated and definitely don't hesitate to have me clarify anything that is unclear.
Many thanks!
Teddy
Before describing my question, I want to thank you and the other CONN folks for providing this wonderful tool and for maintaining this forum. It's a remarkable feat!
Earlier today, I noticed a discrepancy in CONN's seed-to-voxel results explorer and I was hoping to receive some clarification.
I'm noticing that when I start with the effect of group (contrast vector: [1]) in the main GUI and then, in the results explorer, go from the one-sided positive (t) contrast to the two-sided (f) contrast, that the coordinates for the positive clusters stay the same. In contrast, if I compare the coordinates from the one-sided negative (t) contrast to the two-sided (f) contrast, that the coordinates for the negative clusters change.
If I type the F-contrast matrix into the main GUI (e.g., [1; -1] rather than [1]) or use the spm display button to define a new F-contrast, the coordinates are the same as both positive and negative one-sided contrasts. I'm wondering whether the positive one-sided coordinates should be updating when going from one- to two-sided or the negative one-sided coordinates are not updating accurately in the results explorer or I'm misinterpreting in some way. Any guidance you can provide would be greatly appreciated and definitely don't hesitate to have me clarify anything that is unclear.
Many thanks!
Teddy
Jan 10, 2017 05:01 AM | Alfonso Nieto-Castanon - Boston University
RE: seed-to-voxel coordinate for F-contrast
Hi Teddy,
Those differences are likely due to having slightly different clusters in the one-sided vs. two-sided results. You should find that, even for those clusters that still show the same coordinates, their cluster sizes will vary depending on whether you use one-sided vs. two-sided analyses (assuming here that you are using height thresholds based on p-values, e.g. a uncorrected p<.001 height threshold). The coordinates are simply reporting the peak voxel within a cluster, so even if a cluster is slightly larger it may often still have the same peak coordinates. The reason for the different cluster sizes is that the same p-value threshold results in two different T- value thresholds, depending on whether you are using one-sided or two-sided statistics, and these will naturally result in slightly different suprathreshold clusters. To be precise, if you are using a p<.001 height threshold in your two-sided analyses, and you then want to look at one-sided results and make sure that you are looking at exactly the same clusters, then you should use p<.0005 height thresholds in each of your two one-sided analyses.
Hope this helps
Alfonso
Originally posted by Theodore Turesky:
Those differences are likely due to having slightly different clusters in the one-sided vs. two-sided results. You should find that, even for those clusters that still show the same coordinates, their cluster sizes will vary depending on whether you use one-sided vs. two-sided analyses (assuming here that you are using height thresholds based on p-values, e.g. a uncorrected p<.001 height threshold). The coordinates are simply reporting the peak voxel within a cluster, so even if a cluster is slightly larger it may often still have the same peak coordinates. The reason for the different cluster sizes is that the same p-value threshold results in two different T- value thresholds, depending on whether you are using one-sided or two-sided statistics, and these will naturally result in slightly different suprathreshold clusters. To be precise, if you are using a p<.001 height threshold in your two-sided analyses, and you then want to look at one-sided results and make sure that you are looking at exactly the same clusters, then you should use p<.0005 height thresholds in each of your two one-sided analyses.
Hope this helps
Alfonso
Originally posted by Theodore Turesky:
Hi Alfonso et al.,
Before describing my question, I want to thank you and the other CONN folks for providing this wonderful tool and for maintaining this forum. It's a remarkable feat!
Earlier today, I noticed a discrepancy in CONN's seed-to-voxel results explorer and I was hoping to receive some clarification.
I'm noticing that when I start with the effect of group (contrast vector: [1]) in the main GUI and then, in the results explorer, go from the one-sided positive (t) contrast to the two-sided (f) contrast, that the coordinates for the positive clusters stay the same. In contrast, if I compare the coordinates from the one-sided negative (t) contrast to the two-sided (f) contrast, that the coordinates for the negative clusters change.
If I type the F-contrast matrix into the main GUI (e.g., [1; -1] rather than [1]) or use the spm display button to define a new F-contrast, the coordinates are the same as both positive and negative one-sided contrasts. I'm wondering whether the positive one-sided coordinates should be updating when going from one- to two-sided or the negative one-sided coordinates are not updating accurately in the results explorer or I'm misinterpreting in some way. Any guidance you can provide would be greatly appreciated and definitely don't hesitate to have me clarify anything that is unclear.
Many thanks!
Teddy
Before describing my question, I want to thank you and the other CONN folks for providing this wonderful tool and for maintaining this forum. It's a remarkable feat!
Earlier today, I noticed a discrepancy in CONN's seed-to-voxel results explorer and I was hoping to receive some clarification.
I'm noticing that when I start with the effect of group (contrast vector: [1]) in the main GUI and then, in the results explorer, go from the one-sided positive (t) contrast to the two-sided (f) contrast, that the coordinates for the positive clusters stay the same. In contrast, if I compare the coordinates from the one-sided negative (t) contrast to the two-sided (f) contrast, that the coordinates for the negative clusters change.
If I type the F-contrast matrix into the main GUI (e.g., [1; -1] rather than [1]) or use the spm display button to define a new F-contrast, the coordinates are the same as both positive and negative one-sided contrasts. I'm wondering whether the positive one-sided coordinates should be updating when going from one- to two-sided or the negative one-sided coordinates are not updating accurately in the results explorer or I'm misinterpreting in some way. Any guidance you can provide would be greatly appreciated and definitely don't hesitate to have me clarify anything that is unclear.
Many thanks!
Teddy
Jan 10, 2017 06:01 PM | Theodore Turesky
RE: seed-to-voxel coordinate for F-contrast
Dear Alfonso,
Many thanks for your prompt reply! The reason for the different cluster sizes absolutely makes sense and I can understand why going from one-sided positive to two-sided would not change the peak coordinates of these clusters. What concerns me is the discrepancy between the positive and negative aspects: i.e., the peak coordinates for one-sided positive and two-sided results are the same, but the peak coordinates for one-sided negative and two-sided are different. In my attachment, I have screenshots of the two-sided result (p-height: .001), positive one-sided (p-height: .0005), and negative one-sided (p-height: .0005). The little red arrows in the coordinate box of the two-sided screenshot indicates positive results. If you compare the coordinates next to these arrows with the coordinates from the one-sided positive coordinates, you will notice that they are identical. Meanwhile, if you compare the coordinates from the two-sided results that are not accompanied by an arrow with the one-sided negative coordinates, you will notice that every coordinate is different.
Many thanks,
Teddy
Many thanks for your prompt reply! The reason for the different cluster sizes absolutely makes sense and I can understand why going from one-sided positive to two-sided would not change the peak coordinates of these clusters. What concerns me is the discrepancy between the positive and negative aspects: i.e., the peak coordinates for one-sided positive and two-sided results are the same, but the peak coordinates for one-sided negative and two-sided are different. In my attachment, I have screenshots of the two-sided result (p-height: .001), positive one-sided (p-height: .0005), and negative one-sided (p-height: .0005). The little red arrows in the coordinate box of the two-sided screenshot indicates positive results. If you compare the coordinates next to these arrows with the coordinates from the one-sided positive coordinates, you will notice that they are identical. Meanwhile, if you compare the coordinates from the two-sided results that are not accompanied by an arrow with the one-sided negative coordinates, you will notice that every coordinate is different.
Many thanks,
Teddy
Jan 11, 2017 01:01 AM | Alfonso Nieto-Castanon - Boston University
RE: seed-to-voxel coordinate for F-contrast
Dear Teddy,
You are exactly right about this issue, sorry I did not realize it in my previous mail. This was in fact a bug in the way negative-effects peak coordinates were being reported when using two-sided analyses. For negative-effect clusters, the correct coordinates were being reported when using the "one-sided negative" option, but not when using the "two-sided" option. I am attaching a patch to fix this issue (this patch is for release 17a, simply copy this file to the conn distribution folder overwriting the file with the same name there). This issue affected both the peak coordinates being reported as well as the peak-level statistics for negative effects when using two-sided analyses (cluster-size values and cluster-level statistics were not affected by this issue, for neither one-sided nor two-sided analyses). Thank you very much for reporting this (and for the follow-up after I failed to notice), and let me know if this seems to be working fine now
Best regards
Alfonso
Originally posted by Theodore Turesky:
You are exactly right about this issue, sorry I did not realize it in my previous mail. This was in fact a bug in the way negative-effects peak coordinates were being reported when using two-sided analyses. For negative-effect clusters, the correct coordinates were being reported when using the "one-sided negative" option, but not when using the "two-sided" option. I am attaching a patch to fix this issue (this patch is for release 17a, simply copy this file to the conn distribution folder overwriting the file with the same name there). This issue affected both the peak coordinates being reported as well as the peak-level statistics for negative effects when using two-sided analyses (cluster-size values and cluster-level statistics were not affected by this issue, for neither one-sided nor two-sided analyses). Thank you very much for reporting this (and for the follow-up after I failed to notice), and let me know if this seems to be working fine now
Best regards
Alfonso
Originally posted by Theodore Turesky:
Dear Alfonso,
Many thanks for your prompt reply! The reason for the different cluster sizes absolutely makes sense and I can understand why going from one-sided positive to two-sided would not change the peak coordinates of these clusters. What concerns me is the discrepancy between the positive and negative aspects: i.e., the peak coordinates for one-sided positive and two-sided results are the same, but the peak coordinates for one-sided negative and two-sided are different. In my attachment, I have screenshots of the two-sided result (p-height: .001), positive one-sided (p-height: .0005), and negative one-sided (p-height: .0005). The little red arrows in the coordinate box of the two-sided screenshot indicates positive results. If you compare the coordinates next to these arrows with the coordinates from the one-sided positive coordinates, you will notice that they are identical. Meanwhile, if you compare the coordinates from the two-sided results that are not accompanied by an arrow with the one-sided negative coordinates, you will notice that every coordinate is different.
Many thanks,
Teddy
Many thanks for your prompt reply! The reason for the different cluster sizes absolutely makes sense and I can understand why going from one-sided positive to two-sided would not change the peak coordinates of these clusters. What concerns me is the discrepancy between the positive and negative aspects: i.e., the peak coordinates for one-sided positive and two-sided results are the same, but the peak coordinates for one-sided negative and two-sided are different. In my attachment, I have screenshots of the two-sided result (p-height: .001), positive one-sided (p-height: .0005), and negative one-sided (p-height: .0005). The little red arrows in the coordinate box of the two-sided screenshot indicates positive results. If you compare the coordinates next to these arrows with the coordinates from the one-sided positive coordinates, you will notice that they are identical. Meanwhile, if you compare the coordinates from the two-sided results that are not accompanied by an arrow with the one-sided negative coordinates, you will notice that every coordinate is different.
Many thanks,
Teddy
Jan 11, 2017 10:01 AM | Theodore Turesky
RE: seed-to-voxel coordinate for F-contrast
Dear Alfonso,
The patch worked beautifully. Also, absolutely no apology necessary. This is a wonderful service you provide and I think everyone on this forum (and probably exponentially more people) are grateful to you for all that you do here.
Again, many thanks!
Teddy
The patch worked beautifully. Also, absolutely no apology necessary. This is a wonderful service you provide and I think everyone on this forum (and probably exponentially more people) are grateful to you for all that you do here.
Again, many thanks!
Teddy