General

 

The batchtool was created out of the needs to apply common neuroimaging pipelines (like SPM, FSL, Freesurfer) on medium sized projects (10-1000 subjects) in a convenient and efficient manner. One of the major objective of NORA's batchtool is to deal with heterogenous data. Typically, files (a.k.a. image series, or any other filecontent) are selected by file patterns, which can iteratively generalized. Errors can be easily tracked by a simple error logging system. For example, it is simple to select a errornous subgroup and rerun a modified job for them, which was corrected for the error. Processing pipelines (batches) are a simple linear series of jobs. Depending on the relationship between individual jobs, they can run serially or in parallel. 

In conclusion, what it provides:

• Convenient selection of subject/study sets to apply certain processing pipelines
• Definitions of inputs via Tags or filepatterns with wildcards
• Composition of processing pipelines based on predefined scripts/jobs (mostly MATLAB), or custom MATLAB/BASH/Python code
• Submission of jobs to a cluster (Slurm/SGE) with direct access to logs and errors

 

The Batchtool Window

Consider Figure 2 below: the subject/studies table on the left is used for the selection of subject/studies on which you want to run your batch.  You can use the filter bars to create the subgroup you want to work on (see Subject/Studies table).  Select  "Batchtool" on the top toolbar (A) to open the batchtool. Figure 2 shows the structure of the Bacthtool window. It allows to compose the batch out of single jobs. Jobs are added from the menu (D). You can save batches (E), which then appear in the batch list (C). To open an overview of currently running jobs open the "Gridstats" window (B) or (H).

Batches are launched for every subjects/study  independently in parallel.  The jobs within a batch run sequentially. You can also choose different running option (see  (G) in Figure 2). Depending on the selection level (subjects or studies), the batches are iterated over subjects or patients
 
Imagine a scenario where you have multiple studies per patient, which have to be linked in some sense. Then, the subject level is appropriate. For example, think of a neuroimaging analysis where you a have a CT study (which contains, e.g. electrode information)  and a MR study (which contains soft tissue anatomical information), or think of a simple longitudinal analysis. Otherwise, if your your studies should all be treated in an equal manner, the study level is appropriate.

 

 

 

Figure 2: Batchtool overview. 

 

The Anatomy of a Job

A job consists of a list of arguments. There are several types of arguments:

• FILE
  All input images/series (or any other type of files) are given as FILE arguments. Usually you give a file pattern instead of an explicit filename. A fIle pattern is a combination of subfolders, filename and wildcards. For example:  t1*/s0__.nii. I refers to all files contained in a folder starting with t1 and whose filename matches "s0___.nii" .  The asterisks (*) is  a placeholder for an arbitrary character sequence,  an underscore "_" for a single character. Internally, the wildcards are the same as for SQL "like" statement (the '*' is replaced by '%'). A FILE argument also includes a reference to a study or patient. Depending on the selection level (subject or study), different "study references" are possible. See below for more about "study references".
• OUT
  A name of a file including the subfolder. No wildcards are allowed here. Depending on the selection level there are again different study references possible.
• PATHOUT
  Same as OUT but refers to foldername instead of a filename.
• NUMERIC
• STRING
• LOGICAL
• OPTION

 

Study References and study selectors

 

 

Figure 3: The anatomy of a single job.

 

 

Generic jobs

There are a multitude of predefined algorithms (mostly MATLAB) in NORA; however you can also implement your own scripts directly by using generic jobs. Currently there are three types of languages possible:

• BASH
• Python
• MATLAB

 For BASH/Python scripts input files (and all other parameters) are referenced by variables with a $-prefix with a special naming convention. For example, file arguments are referenced by $f1-$f9. Once NORA finds such an expression it automatically adds a corresponding row at the bottom of the job, which can be filled by the appropriate file patterns. The same holds of output arguments (represented by $o1-$o9) and output paths (prefix 'p'). Other parameters (STRING,NUMERIC) are referenced by prefixes 's' and 'n'.

Figure 4: Generic Jobs

 

Cluster Managment/Monitor (Gridstats)

To monitor ....