Pipeline Processing

Basic Concept

The PMOD batch pipeline supports fully automatic image processing. A pipeline is configured as a series of processing steps, each receiving the results of the prior step, performing an additional analysis, and forwarding the results to the next step, with the last step producing the end results. A pipeline definition can be applied to a whole set of input images for obtaining objective results in a group analysis. To inspect intermediate results, a quality control (QC) breakpoint can be enabled. Furthermore, these auxiliary results may be saved for closer interactive inspection.

Use of Pipelines with DICOM Server

Toolbox based DICOM servers are able to execute a specific processing pipeline when they receive images. This feature allows setting up special purpose toolbox DICOM server which perform simple and well-defined processing steps in the background.

Note: Pipelines cannot be used by DICOM servers started from a command script.

Assumptions and Recommendations

PMOD includes a powerful database for the storage and organization of all kinds of relevant data. It is highly recommended that the data for pipeline processing is organized in such a database, because some processing tools will take advantage from specific relationships among data elements which can only be provided by a database. Furthermore, the output produced by the pipeline can directly be inserted at the appropriate database level, which is more difficult when working with directory structures.

A study may comprise data of several groups (e.g. controls, subjects) and conditions (baseline, follow-up). In this case a separate pipeline for each group/condition pair should be prepared as explained below. All pipelines can then be submitted for processing at once, and the results will be conveniently aggregated for group statistics.

Pipeline Elements

A processing pipeline can easily be set up via the interactive user interface explained below. It allows choosing and configuring an arbitrary number of processing steps from a list including

▪the external tools which are available in every PMOD license;

▪automatic brain VOIs, image registration, normalization and motion correction which are tied to the fusion option (PFUS);

▪kinetic modeling of regional TACs which belongs to the modeling option (PKIN), and some

▪pipeline controlling elements.

The pipeline definitions can be saved for later retrieval and processing.

Use Cases of Pipeline Processing

This documentation includes pipeline processing examples representing practical scenarios. Please make sure to read the first example as it provides a detailed step-by-step explanation, which is not repeated in the other examples. Furthermore, the Demo database of the PMOD distribution includes working examples which can be retrieved with Load Processing and directly executed.