Performance-aware programming for intraoperative intensity-based image registration on graphics processing units
By
Martin C. W. Leong
Kit-Hang Lee
Bowen P. Y. Kwan
Yui-Lun Ng
Zhiyu Liu
Nassir Navab
Wayne Luk
Ka-Wai Kwok
January 23, 2021
Clinical Scorecard: Optimizing Programming for Intensity-Based Image Registration During Surgery Using Graphics Processing Units
At a Glance
Category Detail
Condition Non-rigid image misalignment during surgery due to tissue deformation and motion Key Mechanisms Intensity-based non-rigid image registration using diffeomorphic log-demons algorithm accelerated by GPU performance-aware programming Target Population Patients undergoing image-guided interventions, especially MRI-guided cardiac electrophysiology procedures Care Setting Intraoperative surgical environment requiring rapid image registration
Key Highlights
Non-rigid image registration aligns preoperative and intraoperative images despite tissue deformation and motion. Intensity-based registration methods like diffeomorphic log-demons are robust but computationally intensive. GPU-based performance-aware programming significantly reduces registration time to meet intraoperative time constraints. Guideline-Based Recommendations
Diagnosis
Use intensity-based non-rigid registration to correct misalignment between pre-op and intra-op images. Management
Implement GPU-accelerated diffeomorphic log-demons algorithm for rapid intraoperative image registration. Apply performance-aware programming techniques to optimize GPU resource utilization and minimize runtime. Monitoring & Follow-up
Ensure registration time is under 10 seconds post-ablation to integrate seamlessly into surgical workflow. Continuously profile and optimize bottleneck operations during algorithm development. Risks
Delayed registration may impair real-time surgical guidance and affect procedural outcomes. Inaccurate registration due to noise and artifacts if geometry-based methods are used instead of intensity-based. Patient & Prescribing Data
Patients undergoing image-guided surgeries with soft tissue deformation, e.g., cardiac electrophysiology ablations.
Rapid, accurate intraoperative image registration enhances surgical roadmap integration and procedural success.
Clinical Best Practices
Prefer intensity-based registration algorithms for robustness against noise and deformation. Utilize GPU hardware and performance-aware programming to achieve clinically acceptable registration times. Benchmark and optimize algorithm bottlenecks iteratively to meet intraoperative timing requirements. Open-source GPU implementations can facilitate wider adoption and further optimization. References
