FPGA-Accelerated Deformable Image Registration for Improved Target-Delineation During CT-Guided Interventions.

Area of Science:

• Medical Imaging
• Computer Engineering
• Interventional Radiology

Background:

• Minimally invasive image-guided interventions (IGIs) benefit from accurate image registration for precise targeting.
• Nonrigid tissue motion necessitates deformable registration, but existing methods are too slow for clinical use.
• Current rigid registration methods fail to account for anatomical changes between pre- and intraprocedural imaging.

Purpose of the Study:

• To develop and evaluate a hardware-accelerated solution for mutual information (MI)-based deformable image registration.
• To reduce the execution time of deformable registration for integration into the IGI workflow.
• To assess the accuracy and feasibility of the proposed solution for abdominal imaging registration.

Main Methods:

• Implementation of a field-programmable gate array (FPGA)-based architecture for MI-based deformable image registration.
• Testing the system with abdominal intraprocedural noncontrast CT (iCT) images registered against preprocedural contrast-enhanced CT (preCT) and PET images.
• Evaluation of registration accuracy using 5 iCT-preCT and 5 iCT-PET image pairs.

Main Results:

• The FPGA implementation reduced deformable registration time from hours to minutes.
• Registration accuracy was comparable to software-based methods, achieving results on the order of a few millimeters.
• Successful registration of iCT with preCT and PET images was demonstrated.

Conclusions:

• The hardware-accelerated deformable registration system is significantly faster than software-based approaches.
• The achieved accuracy and speed make the solution suitable for real-time integration into IGI workflows.
• This advancement can enhance target delineation and improve the efficiency of image-guided interventions.