CASToR: a generic data organization and processing code framework for multi-modal and multi-dimensional tomographic...

Area of Science:

• Medical Imaging
• Computational Science
• Software Engineering

Background:

• Tomographic reconstruction software development is hindered by data acquisition and organizational differences across modalities (PET, SPECT, CT).
• Existing reconstruction algorithms often have limited applicability, leading to code duplication and difficulties in comparing methods.

Purpose of the Study:

• To propose a unified, generic code framework for formatting, processing, and reconstructing multi-modal and multi-dimensional tomographic data.
• To address challenges in software development, algorithm implementation, and cross-method comparison in tomographic reconstruction.

Main Methods:

• Developed an iterative framework processing both list-mode and histogrammed data uniformly.
• Implemented a generic algorithm engine with abstract components and modality/algorithm-specific extensions.
• Integrated temporal dimension handling and parallel execution capabilities.
• The framework, CASToR, is an open-source C++ parallel platform.

Main Results:

• The generic framework demonstrates acceptable time loss (one order of magnitude slower than optimized manufacturer software).
• Specific optimizations yielded an average speed-up factor of 1.54 to 3.07 compared to conventional implementations.
• Parallel programming achieved near-linear speed-up (0.85x per core) in clinical PET settings.

Conclusions:

• The proposed framework offers significant flexibility for integrating new reconstruction algorithms in medical imaging.
• It maintains computational efficiency, enabling easier comparison and development of tomographic reconstruction methods.
• CASToR facilitates advanced, parallelized tomographic reconstruction across various modalities.