Need for ‘Automatisation’ in Imaging analysis

Remember the last time you were sitting in front of your system with tons of unanalyzed data looking back at you. And remember the kind of motivation you needed to get yourself started on the unintelligent task of button pressing repeatedly, doing the same thing over and over again for all your participants. 

Above statements might sound very familiar to researchers in any field but especially to researchers in Neuro-Imaging where a 20 min MRI scan may produce data as huge as 1 GB! And, where a common practice is to use atleast 16-25 participants for one study. So you are stuck behind your computer, repeating a procedure 16-25 times without you getting any smarter.

All this can be changed by just automatizing the whole process. So that you don’t have to do the entire task manually but the computer does it automatically for you. You just specify the procedure and then let the computer do everything else. How so much magical it might sound, it’s a very feasible option and is practiced by many labs. Customized Batch scripts as they are called are very useful option while analyzing data.

Some of the instances in Neuro-Imaging where making a batch script is useful include 

    1.      fMRI preprocessing and statistics.

    2.      Behavioral data analysis.

fMRI preprocessing and statistics is a very laborious but essential step to be repeated for all participants. SPM8 version of the SPM software package includes a very efficient batch system (Details shown in spm8 manual). Also, the new batch system in spm8 allows for addition of new toolbox by adding config files. These config files can be saved in spm8 toolbox folder and they will be loaded just as other toolboxes (see some sample config files in folders spm8/toolbox or in spm8/config). The new batch system allows you to specify inputs to your function at the same time and an option for dependency (instead of virtual files in spm5).

Behavioral data analysis can be automated in different softwares. One of the easiest ways is to automate it in Microsoft Excel by the use of Macros. Macros is a visual basic addition to the Excel. It’s a set of stored series of commands and functions that can be run to automate a process (see http://office.microsoft.com/en-us/excel-help/CH010100157.aspx). The way you create a macro is by recording your actions just once and it saves whatever you do. This you can do repeatedly for similar data.

For example, a macro code (below) checks whether the accuracy in a list of trials is 1 or not. What it essentially checks is the following condition in Excel : =IF((AND(F4="",E4="")),"",IF(OR(F4=1,E4=1),1,0))

Sub AccuracyRTCorrec()

'

' AccuracyRTCorrec Macro

'

' Keyboard Shortcut: Ctrl+q

'

    Range("M2").Select

    ActiveCell.FormulaR1C1 = "Accuracy"

    Range("M4").Select

    ActiveCell.FormulaR1C1 = _

        "=IF((AND(RC[-7]="""",RC[-8]="""")),"""",IF(OR(RC[-7]=1,RC[-8]=1),1,0))"

    Range("M4").Select

     Selection.Copy

    Range("M5:M162").Select

    ActiveSheet.Paste  

End Sub

The condition checks if the cells F4 and E4 are empty, returns an empty cell for it or else if either one of them is 1, places the value as 1.

Such macros are useful while analyzing behavioral data for different participants and can do the calculation for 80-100 trials in just a second, by pressing a shortcut key for the macro.

Automation not only saves a huge amount of time for the researcher which he/she can use to plan better experiments etc. , it also ensures that there is no human error involved especially when its late in the night and you are in a hurry to analyze data due for a presentation next morning (researchers who have to share facilities and thus have to use them only at night, can relate to this). Plus all of your data goes through the same code unbiased.

Limitations of Automatisation:

As always there are limitations to this entire nice magical seeming thing:

1. Everything can’t be automatised and if a single error creeps in the code, it’s most likely to go unnoticed if not checked properly: So always check the results for anything spurious.

2.  Bad data è Wrong results. But if it’s automated, you may miss it: Keep checking your unanalyzed data for image artifacts and other problems.

3.  You may miss out on interesting trends. Many researchers will agree that data sometimes shows some interesting and surprising trends that are best noticed when you are manually performing the analysis. But if such trends do indeed exist and is not a ghost, then they will survive.

So all in all, automatisation is good. I hope we get more and more customized batch scripts in future…