Thesis
Distributed processing and sSegmentation of medical images with the pydra workflow engine
University of Iowa
Master of Science (MS), University of Iowa
Summer 2021
DOI: 10.17077/etd.005958
Abstract
Neuroscience identifies structural brain patterns that correlate to human characteristics and disease. Medical image analysis of Magnetic Resonance Imaging (MRI), Positron Emission Tomography (PET), or Computed Tomography (CT) provides extensive data points of brain composition. Processing brain scans often connects medical imaging steps, including, among others, intensity normalization, identifying landmarks, inhomogeneity correction, tissue segmentation, regional segmentation, cortical depth segmentation, and finally, resampling and registering to shared space. Preparing brain scan data must be reliably reproducible and scalable to thousands of data sets to contribute to further discovery through medical image analysis.
This project defines a modular pipeline to execute a series of medical imaging software applications to prepare brain scans for further analysis. The SINAPSE Lab at the University of Iowa previously created an automated brain scan analysis tool named AutoWorkup to process raw MRI data. AutoWorkup executes a series of command-line calls to executable medical imaging tools. However, the limited caching ability of Nipype hinders sharing the results of executing these commands. AutoWorkup uses the dataflow engine Nipype to integrate with the grid computing software system Sun Grid Engine (SGE). This SGE interface does not scale well to large datasets as AutoWorkup submits each task as an individual job, exceeds job submission limits, and overloads SGE.
This project recreated AutoWorkup to seamlessly scale to larger datasets and share reproducible results using the dataflow engine Pydra. This work contributed an extension to the Pydra workflow engine that allows for the collection and submission of tasks via job array for scalable distributed computing to avoid exceeding job limits and overloading. Pydra also implements caching entire workflows and sharing a cache across workflows. These features enable sharing AutoWorkup results to be more reproducible and verifiable by the greater medical imaging community. AutoWorkup was recreated with Pydra to take advantage of the updated features of this new dataflow engine.
Details
- Title: Subtitle
- Distributed processing and sSegmentation of medical images with the pydra workflow engine
- Creators
- Charles Edward Johnson III
- Contributors
- Hans Johnson (Advisor)Guadalupe Canahuate (Committee Member)Tyler Bell (Committee Member)
- Resource Type
- Thesis
- Degree Awarded
- Master of Science (MS), University of Iowa
- Degree in
- Electrical and Computer Engineering
- Date degree season
- Summer 2021
- DOI
- 10.17077/etd.005958
- Publisher
- University of Iowa
- Number of pages
- xi, 86 pages
- Copyright
- Copyright 2021 Charles Edward Johnson III
- Language
- English
- Description illustrations
- color illustrations
- Description bibliographic
- Includes bibliographical references (pages 48-49).
- Public Abstract (ETD)
Neuroscience identifies structural brain patterns that correlate to human characteristics and disease. However, processing many brain scans for this analysis at one time is a complex task. Performing this analysis generally involves executing a sequence of coordinated medical imaging steps. Preparing brain scan data must be reliably reproducible and scalable to thousands of data sets to contribute to further discovery through medical image analysis.
This project develops a software application to execute the medical imaging steps to prepare brain scans for further analysis. The SINAPSE Lab previously created a similar application called AutoWorkup using the Nipype framework to process brain scans. However, Nipype has shortcomings in sharing reproducible results and scaling to handle large sets of brain scans.
This project recreated AutoWorkup to seamlessly scale to larger datasets and share reproducible results using the dataflow engine Pydra. This work enhanced Pydra to handle larger datasets by interacting with distributed computing systems in a more scalable way. Pydra also improves upon the Nipype caching implementation to enable sharing reproducible results. This project recreated AutoWorkup with Pydra to take advantage of the updated features of this new dataflow engine.
- Academic Unit
- Electrical and Computer Engineering
- Record Identifier
- 9984124470202771
Metrics
1 File views/ downloads
27 Record Views