processing-scripts > compare templates with the DR results
Showing 1-3 of 3 posts
Jun 18, 2010 02:06 PM | Ningqing Liang
compare templates with the DR results
Dear Experts,
First, thank you for the easy using scripts. I used them to generate individual dual regression maps for data of several centers.
My question is, when I compare the resulted maps with the templates(metaICA.nii.gz, 20 maps included in the script) used, I find they are not as similar to the templates as I expected.
The comparing: dr_ic1 to the first volumn in metaICA and so on.
The criteria for comparing is: find the top 25% of the voxels in dr_ic1 and in the corresponding template, name them as top_active_number_of_map1 and top_active_number_of_map2.
The metric: dice_coefficient = 2* number_of_overlapping_voxels / (top_active_number_of_map1 + top_active_number_of_map2).
Results: the 20 dice_coefficients are mostly between 0.40 and 0.50, and consistently over 100 subjects from group Taipei_a, Beijing_Zang, NewYork_a, Newark and etc. An exception is the fourth map, dr_ic4
with the 4th volumn of templates, the dice_coefficient is at a level between 0.20 and 0.30, and also consistently happens over those subjects.
What I want to ask: what's your expected value of the comparing metric and why? I was expecting 0.75 to 0.95 because the individual maps were regressed using the templates and further more, those
templates were obtained as a consistent componet among the data set of 17*18=306 subjects.
With further investigations to the values of the individual maps, I find that while a voxel in the templates has a positive value, the corresponding voxel of the individual map could be negtive and it is
not rare. The difference (e.x. 1.5 of template but -2.1 of individual map) is beyond my expectation. What's your opinion of this situation?
Here are dice coefficients of the 20 maps of four sample subjects from center Newark, as an example,
0.41 0.38 0.54 0.28 0.46 0.43 0.42 0.50 0.53 0.40 0.57 0.34 0.56 0.50 0.41 0.51 0.48 0.45 0.48 0.51
0.41 0.41 0.44 0.23 0.41 0.40 0.35 0.57 0.50 0.40 0.56 0.37 0.52 0.50 0.40 0.44 0.50 0.50 0.43 0.46
0.38 0.36 0.46 0.22 0.44 0.42 0.33 0.57 0.48 0.38 0.55 0.42 0.54 0.48 0.45 0.49 0.42 0.39 0.41 0.44
0.40 0.35 0.52 0.27 0.36 0.43 0.47 0.54 0.48 0.43 0.54 0.43 0.55 0.45 0.45 0.52 0.52 0.42 0.53 0.52.
and four sample subjects from Beijing_Zang
sub00440 0.50 0.41 0.55 0.28 0.47 0.46 0.44 0.54 0.50 0.41 0.58 0.41 0.55 0.52 0.45 0.55 0.48 0.52 0.47 0.55
sub01018 0.45 0.30 0.43 0.26 0.43 0.49 0.38 0.56 0.51 0.34 0.55 0.45 0.57 0.52 0.48 0.48 0.51 0.51 0.50 0.47
sub01244 0.44 0.44 0.51 0.26 0.43 0.46 0.47 0.52 0.48 0.48 0.56 0.41 0.51 0.49 0.53 0.55 0.48 0.44 0.51 0.51
sub02403 0.42 0.41 0.48 0.32 0.42 0.44 0.49 0.50 0.50 0.43 0.56 0.41 0.53 0.48 0.48 0.53 0.50 0.50 0.47 0.50
The values are of a consistent level, which means the templates has similar effect on each subject.
further, the dice coefficient between each subject are also of this level, the following 3 rows are the dice coefficients of sub00440 with sub01018, sub 01244 and sub02403 of Beijing_Zang, respectively.
Which might provide you some information.
0.39 0.37 0.46 0.42 0.48 0.48 0.38 0.56 0.61 0.41 0.52 0.44 0.54 0.47 0.56 0.51 0.46 0.55 0.51 0.49
0.51 0.52 0.57 0.35 0.48 0.46 0.47 0.56 0.54 0.38 0.55 0.45 0.55 0.44 0.61 0.53 0.50 0.44 0.54 0.55
0.50 0.50 0.46 0.45 0.43 0.56 0.33 0.54 0.59 0.49 0.54 0.45 0.56 0.55 0.60 0.57 0.53 0.51 0.61 0.52.
As I am still a student in this field, my opinions might be misleading.
I realy appreciate any feedback from you. Thank you very much.
Regards,
Ningqing Liang
Eindhoven University of Technology
the Netherlands
First, thank you for the easy using scripts. I used them to generate individual dual regression maps for data of several centers.
My question is, when I compare the resulted maps with the templates(metaICA.nii.gz, 20 maps included in the script) used, I find they are not as similar to the templates as I expected.
The comparing: dr_ic1 to the first volumn in metaICA and so on.
The criteria for comparing is: find the top 25% of the voxels in dr_ic1 and in the corresponding template, name them as top_active_number_of_map1 and top_active_number_of_map2.
The metric: dice_coefficient = 2* number_of_overlapping_voxels / (top_active_number_of_map1 + top_active_number_of_map2).
Results: the 20 dice_coefficients are mostly between 0.40 and 0.50, and consistently over 100 subjects from group Taipei_a, Beijing_Zang, NewYork_a, Newark and etc. An exception is the fourth map, dr_ic4
with the 4th volumn of templates, the dice_coefficient is at a level between 0.20 and 0.30, and also consistently happens over those subjects.
What I want to ask: what's your expected value of the comparing metric and why? I was expecting 0.75 to 0.95 because the individual maps were regressed using the templates and further more, those
templates were obtained as a consistent componet among the data set of 17*18=306 subjects.
With further investigations to the values of the individual maps, I find that while a voxel in the templates has a positive value, the corresponding voxel of the individual map could be negtive and it is
not rare. The difference (e.x. 1.5 of template but -2.1 of individual map) is beyond my expectation. What's your opinion of this situation?
Here are dice coefficients of the 20 maps of four sample subjects from center Newark, as an example,
0.41 0.38 0.54 0.28 0.46 0.43 0.42 0.50 0.53 0.40 0.57 0.34 0.56 0.50 0.41 0.51 0.48 0.45 0.48 0.51
0.41 0.41 0.44 0.23 0.41 0.40 0.35 0.57 0.50 0.40 0.56 0.37 0.52 0.50 0.40 0.44 0.50 0.50 0.43 0.46
0.38 0.36 0.46 0.22 0.44 0.42 0.33 0.57 0.48 0.38 0.55 0.42 0.54 0.48 0.45 0.49 0.42 0.39 0.41 0.44
0.40 0.35 0.52 0.27 0.36 0.43 0.47 0.54 0.48 0.43 0.54 0.43 0.55 0.45 0.45 0.52 0.52 0.42 0.53 0.52.
and four sample subjects from Beijing_Zang
sub00440 0.50 0.41 0.55 0.28 0.47 0.46 0.44 0.54 0.50 0.41 0.58 0.41 0.55 0.52 0.45 0.55 0.48 0.52 0.47 0.55
sub01018 0.45 0.30 0.43 0.26 0.43 0.49 0.38 0.56 0.51 0.34 0.55 0.45 0.57 0.52 0.48 0.48 0.51 0.51 0.50 0.47
sub01244 0.44 0.44 0.51 0.26 0.43 0.46 0.47 0.52 0.48 0.48 0.56 0.41 0.51 0.49 0.53 0.55 0.48 0.44 0.51 0.51
sub02403 0.42 0.41 0.48 0.32 0.42 0.44 0.49 0.50 0.50 0.43 0.56 0.41 0.53 0.48 0.48 0.53 0.50 0.50 0.47 0.50
The values are of a consistent level, which means the templates has similar effect on each subject.
further, the dice coefficient between each subject are also of this level, the following 3 rows are the dice coefficients of sub00440 with sub01018, sub 01244 and sub02403 of Beijing_Zang, respectively.
Which might provide you some information.
0.39 0.37 0.46 0.42 0.48 0.48 0.38 0.56 0.61 0.41 0.52 0.44 0.54 0.47 0.56 0.51 0.46 0.55 0.51 0.49
0.51 0.52 0.57 0.35 0.48 0.46 0.47 0.56 0.54 0.38 0.55 0.45 0.55 0.44 0.61 0.53 0.50 0.44 0.54 0.55
0.50 0.50 0.46 0.45 0.43 0.56 0.33 0.54 0.59 0.49 0.54 0.45 0.56 0.55 0.60 0.57 0.53 0.51 0.61 0.52.
As I am still a student in this field, my opinions might be misleading.
I realy appreciate any feedback from you. Thank you very much.
Regards,
Ningqing Liang
Eindhoven University of Technology
the Netherlands
Jun 18, 2010 02:06 PM | Ningqing Liang
RE: compare templates with the DR results
To make the questions clear:
1, what's your expected value of the comparing metric and why? (In my opinion, it is too low)
2, The differences (e.x. 1.5 of template but -2.1 of individual map) of many voxels are beyond my expectation. What's your opinion of this situation?
3, What has happened to the fourth template that it resulted in such different lower values?
Regards,
Ningqing Liang
1, what's your expected value of the comparing metric and why? (In my opinion, it is too low)
2, The differences (e.x. 1.5 of template but -2.1 of individual map) of many voxels are beyond my expectation. What's your opinion of this situation?
3, What has happened to the fourth template that it resulted in such different lower values?
Regards,
Ningqing Liang
Jun 28, 2010 08:06 AM | Ningqing Liang
RE: compare templates with the DR results
To Xinian Zuo,
Your email has well answered these questions. Thank you very much.
Regards,
Ningqing Liang
Eindhoven University of Technology
Your email has well answered these questions. Thank you very much.
Regards,
Ningqing Liang
Eindhoven University of Technology