Analysis of scalability of high-performance 3D image processing platform for virtual colonoscopy

Area of Science:

• Medical Imaging
• High-Performance Computing
• Computational Science

Background:

• 3D medical imaging demands fast processing for real-time applications.
• High computational power and memory bandwidth are crucial for handling large datasets.
• Existing solutions often struggle with the speed required for interactive medical imaging.

Purpose of the Study:

• To develop a high-performance software platform for 3D medical image processing.
• To enable fast turn-around times in medical imaging applications.
• To leverage affordable commodity computing systems for enhanced performance.

Main Methods:

• Developed the HPC 3D-MIP platform using size-adaptive, distributable block volumes.
• Implemented efficient parallelization strategies for 3D Maximum Intensity Projection (MIP) algorithms.
• Utilized layered parallel software libraries and task scheduling for load balancing.

Main Results:

• Achieved a 12-fold performance improvement in virtual colonoscopy processes on a 12-core workstation.
• Demonstrated the platform's efficiency compared to sequential implementations.
• Showcased potential for high scalability with increasing core counts, analyzed via Amdahl's Law.

Conclusions:

• The HPC 3D-MIP platform effectively accelerates 3D medical image processing.
• Commodity computing systems can be successfully utilized for high-performance medical imaging.
• The platform shows significant promise for real-time and interactive medical imaging applications.