On the Usage of GPUs for Efficient Motion Estimation in Medical Image Sequences

Area of Science:

• Biomedical imaging
• Computer vision
• High-performance computing

Background:

• Biomedical imaging generates vast datasets, demanding efficient computational methods.
• Multicore architectures, particularly Graphics Processing Units (GPUs), offer potential for accelerating image analysis.
• Real-time performance is crucial for many biomedical applications.

Purpose of the Study:

• To investigate the mapping of an enhanced motion estimation algorithm onto novel GPU architectures.
• To identify and address challenges and benefits associated with GPU implementation.
• To evaluate performance gains and explore architectural optimizations.

Main Methods:

• Developed and mapped an enhanced motion estimation algorithm to GPU-specific architectures.
• Utilized a database of three-dimensional (3D) image sequences for testing.
• Analyzed algorithm performance across three different GPU architectures.
• Investigated GPU memory configurations and access patterns.

Main Results:

• Achieved substantial performance improvements, up to a factor of 60.
• Demonstrated the feasibility of near real-time processing for 3D biomedical image sequences.
• Identified specific GPU architectural features that benefit the algorithm's performance.
• Provided a comprehensive analysis of performance across diverse GPU architectures.

Conclusions:

• Mapping enhanced motion estimation algorithms to novel GPU architectures yields significant speedups.
• GPU optimization strategies are crucial for efficient biomedical image analysis.
• The approach enables near real-time performance, advancing clinical and research applications.