(Sunday, 7/12/2020) [Eastern Time (GMT-4)]
Room: AAPM ePoster Library
Purpose: To identify robust radiomics features extracted from phantom and patient images acquired with a low magnetic field (0.35 T) MR-Linac.
Methods: Eleven scans collected over a 13-month period using a Magphan RT phantom, and 11 scans collected over 11 days using a ViewRay Daily QA phantom, constituted the phantom data representing ideal imaging conditions. Images were acquired with a True Fast Imaging with Steady State Free Precession (TRUFI) pulse sequence with 1.5mm3 voxel resolution, 500mm x 449mm x 432mm field-of-view (FOV), and 172s imaging time. The patient dataset included 12 images from two stereotactic body radiation therapy (SBRT) pancreas patients (1 simulation + 5 fractions) acquired with 1.5mm2 x 3.0mm voxel resolution, 540mm x 465mm x 432mm FOV, and 25s imaging time. To ensure heterogeneity, healthy kidneys were treated as the region of interest (n=24). Variability of 1087 radiomic features (48 categories) was assessed with the coefficient of variation (COV); COV<5% being classified as low/acceptable variability or robust.
Results: After statistical analysis, 15 feature categories were identified as displaying a high number of robust features. These included 91 radiomic features demonstrating robustness (COV<5%) among both phantom and patient data. Robust categories were fractal dimension, intensity and shaped, wavelets (LHL, LLL, HHH, HLL, LHH), Laplacian transforms (sigma=0.5, 1.5, 2.0, 2.5mm), Laws (LLL), Run-length and Co-occurrence based features.
Conclusion: These early results indicate that there are radiomic features demonstrating low/acceptable variability in both phantom and human images obtained with a low-field MR-Linac. We conclude, first, that phantom measurements can be used to identify robust MR radiomic features; and second, that a low-field scanner of a MR-Linac is sufficiently stable over time for radiomics studies. While preliminary, these results hold promise demonstrating low variability of radiomic features extracted from images of a low-field scanner of a MR-Linac.
Funding Support, Disclosures, and Conflict of Interest: RSES was supported by the Swedish Board of Student Finance (CSN) and a Crafoord Foundation Travel Grant.
Keywords
Quantitative Imaging, MRI, Feature Selection
Taxonomy
IM/TH- MRI in Radiation Therapy: MRI/Linear accelerator combined (general)
Contact Email
