Hello everyone,
I have decided to analysis some task mode fMRI data set for training a multi voxel classifier. I want to know what are the most famous feature selection methods for selecting effective voxels after preprocessing the dataset and before applying the data to classifier? I want to say thank you in advance for your kind support.
I would like to refer you to the sections on "2.2. Comparison of Classifiers and Preprocessing Strategies" and "3. Feature Selection and Dimensionality Reduction" in the article
""Computational and Mathematical Methods in Medicine
Volume 2012 (2012), Article ID 961257, 14 pages
http://dx.doi.org/10.1155/2012/961257
Review Article
Multivoxel Pattern Analysis for fMRI Data: A Review
Abdelhak Mahmoudi,1 Sylvain Takerkart,2 Fakhita Regragui,1 Driss Boussaoud,3 and Andrea Brovelli2"".
""https://code.google.com/p/princeton-mvpa-toolbox/wiki/ManualPreClassification ""and ""https://code.google.com/p/princeton-mvpa-toolbox/wiki/TutorialClass"" has also detailed the steps in the evaluation of the model.
I hope it was useful. I learnt a lot through this question myself. From what I could understand, it seems that the type of the training set we use and the question we are trying to answer have a huge bearing on what feature selection method we use, hence being tailored to the respective experiment.
Keenly looking forward to more experienced colleagues guiding us through this question.
Almost all feature selection methods can provide better classification accuracy. But if you don't want to take a lot of time on this case, I think, you shouldn't miss Correlation Based Filters, I have used this method a lot in many situations and I think It really works. It is useful to take a look at Hall's master thesis. you can find it here: http://www.cs.waikato.ac.nz/~mhall/thesis.pdf
Dear Mehdi,
In Brain function analysis, feature selection means selecting a group of voxels for analyzing a special task in the Brain. By contrast of normal feature selection in Machine Learning, here, we must select the parts of Brain e.g. motor cortex, visual cortex, etc., which are affected in our experiment's task(s), anatomically and practically. Further, most of the time, correlation between voxels in brain are dramatically different between different people in different experiments. As a result, we can't apply a usual feature selection on fMRI data set.
- Badges
- Science topic