Characterization of Speed and Accuracy of a Nonrigid Registration Accelerator on Pre- and Intraprocedural Images

Abstract

AbstractTargeting of FDG-avid PET-visible but CT-invisible lesions with CT is, by definition, challenging and is a source of uncertainty in many percutaneous ablative procedures. Preprocedural PET has been overlaid on intraprocedural CT to improve the visualization of such lesions. Such an approach depends on registration of PET and CT, but current registration methods remain mostly slow and manual and do not optimally account for deformation of abdominothoracic anatomy. We have developed a fully automatic, nonrigid image registration technology that is accelerated on 3 field programmable gate array (FPGA) chips to execute in <1 min. We have tested the application of this technology for multimodality PET-CT image guidance during interventional procedures. Archived abdominothoracicpreprocedural PET and intraprocedural CT images of 7 patients with FDG-avid lesions were re-registered using the FPGA method. During the actual procedure, a PET-CT fusion image to assist in needle placement was created for each of the patients using a prototype electromagnetic tool tracking system and manual rigid image registration lasting >5 min. The speed and accuracy of registration, which are critical to eventual clinical adoption, were recorded. On average, the FPGA-based registration took 53 s and agreed with the existing solution at the lesion center to within 6.6 mm (1.7 voxels). We prove the feasibility of fast and accurate nonrigid registration capable of enabling efficient multimodality PET-CT image-guided interventional procedures. The sub-minute speed of the FPGA method is important for clinical efficiency and on-demand intraprocedural PET-CT registration. The accuracy of FPGA-based nonrigid image registration is also acceptable. Our next steps are to introduce the FPGA registration in the clinic and re-test its accuracy, speed, and effectiveness in a larger patient population. When fully developed and tested, our approach might improve target visualization and thus the precision, safety, and outcomes of interventional radiology procedures.KeywordsPositron Emission TomographyPositron Emission Tomography ImageImage RegistrationField Programmable Gate ArrayNonrigid RegistrationThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.