BLUE RIBBON POSTER IMAGING: Dose-Free Functional Lung Imaging: Comparing Deformable Image Registration Algorithms for MR-Only Ventilation Mapping
Abstract
Purpose
Recent advances in MRI could offer patients with benign lung disease whole-lung ventilation mapping that may serve as a surrogate for pathophysiological changes such as increased airway resistance and decreased lung compliance. Ventilation maps can be derived from deformable image registration (DIR) algorithms, but DIRs remain underexplored in MRI lung scans. This work evaluates and compares the registration of MRI lung scans across five DIR algorithms for the purposes of ventilation mapping.
Methods
Free-breathing, non-contrast, 3D lung scans of five participants were acquired on a 1.5T MRI Siemens Magnetom Sola. An ultra-short echo time stack-of-spirals sequence, with reconstructed golden-angle radial sparse parallel and prospective gating, accounted for respiratory motion through acquiring k-space data over inhale and exhale phases. Five open-source DIR algorithms were used to measure exhalation to inhalation deformation vector fields (DVFs): dense displacement sampling (deeds), advanced normalization tools (ANTs), elastix, MATLAB’s demons, and MATLAB’s total variation (TV). Each registration was evaluated following AAPM TG-132 guidelines. The ventilation was characterized using the Jacobian determinate and compared across algorithms using the structural similarity index measure (SSIM).
Results
Aside from MATLAB TV, the DIR algorithms met the tolerances of AAPM TG-132. Deeds and ANTs were the top performing DIR algorithms overall. The ventilation maps varied widely based on the DIR algorithm used, with deeds and elastix producing the most similar ventilation maps at a participant-averaged SSIM of 0.51±0.06.
Conclusion
Though four of five DIR algorithms were TG-132 compliant, the ventilation revealed disparities across all DIR algorithms, necessitating further investigation. Immediate next steps include rigorous DIR algorithm optimization. Future work will also focus on using 4DCT as a benchmark to better understand ventilation in MRI and to explore its clinical relevance as a non-ionizing alternative to regional ventilation imaging.