Image quality vs. statistical power in neuroactivation