Consistent Activation Across Trials and Field Strengths by ROC-Reproducibility Thresholding

Tynan Reid Stevens¹, David B. Clarke², Ryan CN D'Arcy¹, Steven D. Beyea¹

Setting activation thresholds remains a significant challenge in fMRI. Fixed threshold levels do not account for individual variability, habituation, or acquisition strategies. We developed a method for determining data-driven thresholds levels based on test-retest ROC-reproducibility (ROC-r) analysis. Eight subjects performed test-retest fMRI at two field strengths (1.5 & 4 Tesla). Optimized ROC-r thresholds were calculated for the template (5.27 +/- 0.22) and retest images (3.71 +/- 0.18). The ROC-r thresholds were significantly higher for 4 Tesla than 1.5 Tesla. Compared to FDR or Bonferroni thresholds, ROC-r thresholds produced more consistent activation extent across field strengths and trials.