A Omidi¹*, E Weiss², J Wilson^1,3, M Rosu-Bubulac², (1) Biomedical Engineering Department, Virginia Commonwealth University, Richmond, VA, (2) Radiation Oncology Department, Virginia Commonwealth University, Richmond, VA, (3) Pauley Heart Center, School of Medicine, Virginia Commonwealth University, Richmond, VA

• 
  A Omidi

MO-IePD-TRACK 1-2 (Monday, 7/26/2021) 5:30 PM - 6:00 PM [Eastern Time (GMT-4)]

Purpose: The first step in correlating local radiation dose and regional cardiac magnetic resonance (CMR) findings is mapping CT-derived radiation doses onto CMR via image registration. This is the first study to evaluate the accuracy of intra- and inter-modality deformable image registration (DIR) for heart, left ventricle (LV), and thoracic aorta (TA), using paired longitudinal CT and pre/post contrast MR imaging.

Methods: 0% (INH) and 50% (EXH) breathing-phases from 4DCT, and inspiratory/expiratory pre-/post-contrast MR image sets were acquired from five lung cancer patients twice during radiotherapy treatment. DIR was performed for the following comparisons: Different time points - (C1) intramodal CT, (C2) intramodal MR, (C3) intermodal CT/MRI; Same time point - (C4) intermodal CT/MR, (C5) intramodal MR pre/post contrast. Registration performance was assessed using HD, MDA, and DICE for the original and deformed contours.

Results: Mean MDA ranged 1-2 mm, except for the two intermodality comparisons of the LV (2.6mm for both). Mean HD ranged 8-17 mm. Mean DICE ranged 0.85-0.95. Overall, all metrics performed better intramodality, with/without contrast. Contour comparisons via HD and MDA were least favorable for LV, and similar for TA and heart. DICE was best in heart, with LV and TA performing similarly. The following scenarios had statistically significant differences (p<0.05): HD: (C2), (C4), (C5) for LV_vs._TA; MDA: (C1), (C2), (C4), (C5) LV_vs._heart (C2)-(C5) LV_vs._TA; DICE: (C1)-(C5) for TA_vs._heart and LV_vs._heart. When the various modalities were compared, the differences were found significant mainly between (C1) vs. (C4), and (C2) vs. (C4), for almost all structures of interest.

Conclusion: Though intramodal registration remains slightly superior, both intramodal and intermodal DIR are within TG 132 recommended range for the structures of interest, suggesting that quantitative regional analyses of CT and MR images from patients undergoing radiotherapy may be spatially correlated without significant error following registration, even across modalities.

ePosters

Keywords

CT, Registration, MR

Taxonomy

IM/TH- Image Registration: Multi-modality registration

Contact Email