Correlation of distinguishability with other measures

It is interesting to see how well the distinguishability V would correlate with other measures, for instance MI and single voxel accuracy.

On-going experiment

  • Plot V vs single-voxel accuracy
  • Plot V vs MI
  • Plot MI vs accuracy

This is for VTC voxels (click to see the higher resolution figure)

and here is for the whole-brain voxels (click to see the higher resolution figure)

In VTC voxels

  • Pearson's correlation between accuracy and V and MI are 0.64 and 0.69 respectively, the latter is a bit higher. This fact suggests both measures are good for estimating the single-voxel accuracy.
  • The high correlation (0.8) between MI and V suggests the similarity in behavior in both measures when using voxels in VTC.

In whole-brain voxels

  • The correlation between accuracy and V and MI are 0.5 and 0.43 respectively. In this case, V is better than MI for explaining linear relationship with accuracy
  • The correlation between V and MI is reduced to 0.61
  • There are lots of voxels gives as high MI as 0.3 whereas the V value is as low as 0.05; this is a disadvantage of using MI

MATLAB code is made available

  • In order to calculate the accuracy for each voxel, you can use CalculateAccuracySingleVoxel.m
  • To plot the correlation between the measures, use CompareDistinguishabilityVSAccuracy.m