Handling Big Data in Medical Imaging: Iterative Reconstruction with Large-Scale Automated Parallel Computation

Area of Science:

• Medical Imaging
• Computational Science
• Parallel Computing

Background:

• Dynamic cardiac single photon emission computed tomography (SPECT) requires computationally intensive image reconstruction.
• The Maximum Likelihood Expectation Maximum (MLEM) algorithm is a standard iterative method for SPECT reconstruction.
• Current MLEM implementations can be slow, limiting clinical applicability for dynamic studies.

Purpose of the Study:

• To implement the MLEM algorithm for dynamic cardiac SPECT on the Spark/GraphX parallel computing platform.
• To leverage Spark/GraphX for large-scale parallel processing of SPECT data.
• To demonstrate performance improvements and potential clinical usability of the parallelized MLEM algorithm.

Main Methods:

• Porting the MLEM algorithm to Spark/GraphX, a parallel computing system.
• Utilizing GraphX for graph and sparse linear algebra operations within Spark.
• Evaluating the performance of the implemented MLEM algorithm on large datasets.

Main Results:

• Successful implementation of the MLEM algorithm on Spark/GraphX.
• Demonstrated performance gains compared to traditional implementations (details to be presented).
• The implementation allows parallelization without requiring specialized parallel computing expertise.

Conclusions:

• Spark/GraphX provides a viable platform for parallelizing MLEM reconstruction for dynamic cardiac SPECT.
• This approach enhances computational efficiency, paving the way for faster image reconstruction.
• The developed method has the potential to be integrated into clinical workflows for improved patient diagnosis.