Accelerating B-spline interpolation on GPUs: Application to medical image registration.

Area of Science:

• Medical Imaging
• Computer-Aided Surgery
• Scientific Computing

Background:

• B-spline interpolation (BSI) is crucial for 3D modeling in medical imaging and Image Guided Surgery (IGS).
• Image registration, essential for IGS, is computationally intensive, hindering real-time applications.
• Current CPU limitations necessitate GPU acceleration for demanding IGS tasks.

Purpose of the Study:

• To develop a novel GPU implementation of BSI to accelerate non-rigid image registration algorithms.
• To optimize BSI for efficient computation in 3D medical imaging applications.
• To enhance the performance and accuracy of Image Guided Surgery workflows.

Main Methods:

• Implemented BSI on GPUs, minimizing data transfer and maximizing on-chip register file usage.
• Reformulated BSI using trilinear interpolations to reduce computational complexity and enhance accuracy.
• Integrated the optimized BSI into a liver deformation registration workflow for pre-clinical validation.

Main Results:

• Achieved an average performance improvement of 6.5x and a 2x increase in interpolation accuracy compared to state-of-the-art GPU implementations.
• Demonstrated up to a 34% acceleration of a non-rigid image registration algorithm (Free Form Deformation) through pre-clinical validation.
• Validated the effectiveness of the GPU-accelerated BSI in a practical medical imaging registration scenario.

Conclusions:

• The novel parallelization scheme effectively utilizes GPU resources, yielding significant performance and accuracy gains.
• The optimized BSI implementation demonstrably enhances the performance of current medical imaging registration applications.
• This GPU-accelerated BSI is a valuable advancement for real-time Image Guided Surgery and other medical imaging tasks.