Cone-beam CT (CBCT) during Y90 radioembolization (RE) of HCC allows precise selection of tumor-feeding arteries and improves accuracy of dose delivery. Co-registration with prior MRI provides additional guidance, particularly in poorly enhancing lesions. Clinically available rigid co-registration algorithms cannot account for anatomy deformations from respiratory motion and require manual interaction. Acceptable co-registration quality is unpredictable and user-dependent, limiting adoption of CBCT-MRI fusion into clinical practice. An automatic, non-rigid co-registration algorithm using the open source software packages ANTs, SLICER and Python was developed for fusion of preprocedural T1 late arterial phase MRI and CBCT acquired during the mapping phase of RE. Co-registration was performed in seven treatment-naive patients with ≤2 lesions measuring 2-10 cm. Clinically meaningful tumor and total liver volume overlays were assessed by two IR attending physicians on a nominal scale (0-5) and analyzed by the Wilcoxon test. Quantitative assessment of liver volume overlay by the non-rigid approach was performed using the dice similarity measure. Average scores for clinically meaningful overlay of target tumor(s) (4.3 vs. 4.3, p>0.99) and liver volumes (4.5 vs. 4.5, p>0.99) were identical for rigid and non-rigid algorithms. Dice similarity measure of liver volumes following non-rigid CBCT-MRI fusion was 0.88 ± 0.04. This is comparable to the best-reported measures in the literature for multi-modal, non-rigid liver co-registration. Clinically relevant CBCT-MRI fusion in HCC using an automatic, non-rigid co-registration algorithm from open source software is feasible. With further development, adoption of this approach may improve reliability and accuracy of image guidance for RE.