Respiratory Motion Compensation Algorithm Enhances Automatic Vessel Detection and Improves Cone-beam CT Based Navigation in Liver Directed-therapies

Objective: Cone-Beam CT (CBCT) combined with automatic vessel detection software are increasingly used in liver directed-therapies. The software detection accuracy relies on high image quality and are susceptible to motion artifact. Breath-hold can be challenging for some patients, leading to repeated acquisitions, inaccurate 3D rendering and failure of the automatic vessel detection. We assessed the impact of motion compensation on the software performance in procedure planning and navigation

Methods: An IRB approved retrospective evaluation of liver directed-therapies from 2014 to 2017 was performed. Presence of breathing motion in CBCT using the 2-dimensional rotational images of the acquisition was confirmed by 2 interventional radiologists. Intra-arterial contrast enhanced CBCTs with at least one well-defined tumor were included. A maximum of 3 tumors per study were analyzed. Automatic tumor supplying vessels detection software was processed to reconstructions with and without respiratory motion compensation (Figure). Sensitivity (true positive rate) and specificity (1 – false positive rate) of the automatic vessels detection were measured before and after motion compensation, pre-operative exams and super selective digital subtraction angiography (DSA) were used as gold standard. 
The ability of using the vicinity vasculature extracted by the software to guide the procedure was also assessed. Finally, the virtual injection feature was tested among the true positive tumor supplying vessels and its accuracy was controlled. Wilcoxon signed rank test was used to compare the sensitivities and specificities.

Results: Motion compensation algorithm was applied retrospectively to 18 CBCTs performed for liver directed-therapies guidance. A total of 30 tumors were analyzed. In 8 (44%) CBCTs the automatic detection failed to find any supplying vessel before motion compensation. At least one supplying vessel was detected for all the procedures after motion compensation. For 10 CBCTs with automatic vessel detection (22 tumors), sensitivity and specificity of vessel extraction were significantly higher after motion correction (p < 0.002 and p < 0.0001 respectively) (Table). For centrally located lesions (Segments I, IV, and V), there was no significant difference in sensitivity (p = 0.06) after motion compensation (Table). The ability of using the tumor supplying vasculature model extracted by the software to guide the procedure was verified in 28 (93%) tumors after motion compensation vs 10 (33%) tumors before. For the positive supplying vessels, the virtual injection was accurate in 84% of the vessels vs 43 % before.

Conclusions: Respiratory motion compensation algorithm improves automatic vessel detection performance and liver vasculature vicinity for CBCT acquisitions exhibiting breath motion artifacts.