DOI: 10.1007/s00259-018-4040-1Pages: 1-9

Quantitative evaluation of tau PET tracers 18F-THK5351 and 18F-AV-1451 in Alzheimer’s disease with standardized uptake value peak-alignment (SUVP) normalization

1. New York University School of Medicine, Center for Brain Health, Department of Psychiatry

2. Tohoku Medical and Pharmaceutical University, Division of Pharmacology, Faculty of Medicine

3. New York University School of Medicine, Department of Radiology

Correspondence to:
Yi Li
Tel: 1-212-263-3258




Off-target binding in the reference region is a challenge for recent tau tracers 18F-AV-1451 and 18F-THK5351. The conventional standardized uptake value ratio (SUVR) method relies on the average uptake from an unaffected tissue sample, and therefore is susceptible to biases from off-target binding as well as variability among individuals in the reference region. We propose a new method, standardized uptake value peak-alignment (SUVP), to reduce the bias of the SUVR, and improve the quantitative assessment of tau deposition.


The SUVP normalizes uptake values by their mode and standard deviation. Instead of using a reference region, the SUVP derives the contrast from unaffected voxels over the whole brain. Using SUVP and SUVR methods, we evaluated the global and regional tau binding of 18F-THK5351 and 18F-AV-1451 on two independent cohorts (N = 18 and 32, respectively), each with cognitively normal (NL) subjects and Alzheimer’s disease (AD) subjects.


Both tracers showed significantly increased binding for AD in the targeted cortical areas. In the temporal cortex, SUVP had a higher classification success rate (CSR) than SUVR (0.96 vs 0.89 for 18F-THK5351; 0.86 vs 0.75 for 18F-AV-1451), as well as higher specificity under fixed sensitivity around 0.80 (0.70 vs 0.45 specificity for 18F-THK5351; 1.00 vs 0.78 for 18F-AV-1451). In the cerebellar cortex, an AD-NL group difference with effect size (Cohen’s d) of 0.62 was observed for AV-1451, confirming the limitation of the SUVR approach using this region as a reference. A smaller cerebellar effect size (0.09) was observed for THK5351.


The SUVP method reduces the bias of the reference region and improves the NL-AD classification compared to the SUVR approach.

To access the full text, please Sign in

If you have institutional access, please click here

  • Accepted: Apr 19, 2018
  • Online: Apr 27, 2018

Article Tools