GPU-accelerated Block Matching Algorithm for Deformable Registration of Lung CT Images.

Area of Science:

• Medical Imaging
• Computer Vision
• Computational Anatomy

Background:

• Deformable registration (DR) is crucial in medical imaging for aligning images.
• Current DR methods often suffer from long processing times and insufficient accuracy, limiting clinical adoption.
• Lung CT image registration presents unique challenges due to respiratory motion and anatomical variability.

Purpose of the Study:

• To develop a computationally efficient and accurate deformable registration algorithm for lung CT images.
• To leverage parallel processing for accelerating the registration pipeline.
• To improve the clinical applicability of deformable registration techniques in thoracic imaging.

Main Methods:

• A parallel block matching algorithm was developed for lung CT image registration.
• The sum of squared difference metric was adapted as the cost function for registration.
• Moving least squares were employed to reconstruct the dense displacement field.
• The algorithm was implemented using NVIDIA's Compute Unified Device Architecture (CUDA) on Graphics Processing Units (GPUs).

Main Results:

• The proposed parallel block matching method significantly reduced registration runtime.
• The algorithm achieved a high registration accuracy with an average error of 1.08 mm (standard deviation 0.69 mm).
• GPU acceleration demonstrated substantial performance gains compared to traditional CPU-based approaches.

Conclusions:

• The developed parallel block matching algorithm offers a fast and accurate solution for lung CT deformable registration.
• The method addresses the limitations of existing techniques, enhancing potential for clinical translation.
• GPU implementation is effective in accelerating complex image registration tasks in medical applications.