Background We assessed the clinical utility of β-amyloid (Aβ) imaging with ¹⁸F-florbetaben (FBB) in mild cognitive impairment (MCI) by evaluating its prognostic accuracy for progression to Alzheimer's disease (AD), comparing semiquantitative with visual scan assessment, and exploring the relationships among Aβ, hippocampal volume (HV) and memory over time.

Methods 45 MCI underwent FBB positron emission tomography, MRI and neuropsychological assessment at baseline and 2 years and clinical follow-up at 4 years. Positive FBB (FBB+), defined by a cortical to cerebellar cortex standardised uptake value ratio (SUVR) ≥1.45, was compared with visual assessment by five readers. Amnestic MCI (aMCI) was defined by a composite episodic memory (EM) Z-score of <−1.5.

Results At baseline, 24 (53%) MCI were FBB+. Majority reads agreed with SUVR classification (κ 0.96). In 2 years, 18 (75%) FBB+ progressed to AD compared with 2 (9.5%) FBB−, yielding a predictive accuracy of 83% (95% CI 61% to 94%). Four FBB− developed non-AD dementia. Predictive accuracies of HV (58% (95% CI 42% to 73%)) and aMCI status (73% (95% CI 58% to 81%)) were lower. Combinations did not improve accuracy. By 4 years, 21 (87.5%) FBB+ had AD whereas 5 (24%) FBB− had non-AD dementia yielding a predictive accuracy of 94% (95% CI 74% to 99%). While the strong baseline association between FBB SUVR and EM declined over 2 years, the association between EM and HV became stronger. FBB SUVR increased 2.2%/year in FBB+ with no change in FBB−.

Conclusions ¹⁸F-florbetaben Aβ imaging facilitates accurate detection of prodromal AD. As neurodegeneration progresses, and in contrast with the early stages of the disease, hippocampal atrophy and not Aβ, seems to drive memory decline.

Trial registration number NCT01138111.