Limits of uncertainty in measured values of embolus-to-blood ratios in dual-frequency TCD recordings due to nonidentical sample volume shapes.