Repeatability and Reproducibility of Radiomics in Magnetic Resonance Fingerprinting of the Brain
R Delgadillo1, B Rich1, Y Yang2, F Yang1. (1) Department Of Radiation Oncology, University Of Miami, Miami, FL, (2) University of Science and Technology of China, Hefei, China
Presentations
PO-GePV-M-46 (Sunday, 7/25/2021) [Eastern Time (GMT-4)]
Purpose: Magnetic Resonance Fingerprinting (MRF) emerged as a quantitative MRI modality allowing for simultaneous measurement of multiple tissue properties in a single, time-efficient acquisition and offers enormous potential to improve diagnosis and therapeutic efficacy. Owning largely to the novelty of this technique, the feasibility of radiomics for quantitative measurements obtained using MRF has been little studied. This study aimed to assess the repeatability and reproducibility of radiomics in MRF-based relaxation time mapping of the brain.
Methods: The imaging cohort included five healthy volunteers, each with the brain scanned on four MRI systems (two at 3.0T and two at 1.5T, all from a single vendor) located at two sites. MRF data was obtained using 2D steady-state free precession with a spiral readout trajectory. T1 and T2 maps were determined through voxel-wise pattern matching between the signal evolution and simulated dictionary. After registration of MRF-derived T1 and T2 maps to an anatomical atlas, 40 radiomics metrics related to gray-level run length, co-occurrence, zone size, and neighborhood difference were evaluated for each of the resultant parametric maps within one VOI located in the parietal lobe featuring strong spatial heterogeneity. Radiomics feature test/retest repeatability and inter-site reproducibility were measured with within-subject variability (WSV) and coefficient-of-variation (CV), respectively.
Results: The test/retest repeatability of the studied radiomics features varied with mean±SD WSVs of 7.7%±12.1% at 1.5T and 8.2%±12.7% at 3.0T on T1 maps while varying with mean±SD WSVs of 7.9%±11.2% at 1.5T and 10.9%±13.4% at 3.0T on T2 maps. As to the inter-site reproducibility, CV of the studied radiomics features varied with mean±SDs of 7.1%±9.2% at 1.5T and 9.1%±11.9% at 3.0T on T1 maps while varying with mean±SDs of 12.0%±13.1% at 1.5T and 15.8%±18.4% at 3.0T on T2 maps.
Conclusion: Compared to weighted images MRF-derived parametric maps may allow for more repeatable and reproducible radiomics analysis.
Keywords
Not Applicable / None Entered.
Taxonomy
Not Applicable / None Entered.
Contact Email