Demographics of the patients are shown in Table 1. Altogether 34 patients were included in the study (23 men and 11 women). The mean age of the patients was 73.4 ± 6.6 (mean ± SD) at baseline. All patients met the criteria of amnestic MCI [13], and all patients gave their written informed consent, which was obtained according to the requirements of the Declaration of Helsinki. The study protocol was approved by the Ethics Committee of Hospital District of Southwest Finland.

The patients were assessed with 18F-Flutemetamol PET scan, brain MRI and cognitive tests. Cognitive tests were administered yearly until conversion to Probable Alzheimer’s Disease or the end of the 3-year follow-up. Probable Alzheimer’s disease was diagnosed when the patients fulfilled the National Institute of Neurological and Communicative Disorders and Stroke-Alzheimer’s Disease and Related Disorders Association criteria as well as DSM-IV criteria for dementia of the Alzheimer type. MRI and 18F-flutemetamol PET were repeated after the subject had converted to AD or after 3 years had elapsed from the first 18F-Flutemetamol PET scan. 18F-Flutemetamol composite cortical uptake value ratio > 1.4 was used as a cut-off value of amyloid positivity [14]. Patients with amyloid-positive and amyloid-negative 18F-Flutemetamol scans at baseline did not differ regarding education and age at baseline (Table 1). Because the study was ongoing for several years, the PET scanner in the PET Centre changed during the study. Therefore, 14 patients were examined with one and 20 with the other scanner (see below for details). Baseline and follow-up scan was done with the same scanner in all but in 2 patients.

From the original 34 participants, seven dropped out the study after first scan and cognitive tests. Four of those 7 patients had positive baseline 18F-Flutemetamol PET scan. Five patients dropped out due to worsening of clinical condition, one participant died because of acute pulmonary embolism and one withdrew the consent due to personal reasons. Altogether 4 participants converted to AD during the study, 3 of them were amyloid positive and one was negative both at baseline and follow-up.

Additional 2 patients were excluded from the PET-analyses due to technical problems in image acquisition. Thus, both baseline and follow-up 18F-Flutemetamol PET data were available from 25 participants.

All patients underwent a 18F-Flutemetamol PET scan. They received approximately 185 MBq of intravenous 18F-Flutemetamol (187.8 ± 18.7 MBq, mean ± SD) and 90 min later underwent a 30-min brain scan. PET scan was done at baseline and approximately after 3-year follow-up or earlier if the patient converted to AD. The mean 18F-Flutemetamol scanning interval was 2.9 ± 0.4 years. Two patients’ follow-up PET scans were excluded from the analysis due to inadequate cerebellar imaging precluding reliable determination of reference region; one had positive and the other negative amyloid status at baseline.

First, the ECAT EXACT HR + (CTI/Siemens, Knoxville, TN, US) was used for PET imaging in 3D mode. The scanner has an axial field of view of 15.5 cm and a patient port of 56.2 cm, and physical performance evaluations of the scanner have shown radial and tangential average spatial resolution of 4.39 mm FWHM and axial resolution of 5.10 mm FWHM (16).

Later, both PET scans and follow-up scan in two patients, which were examined with ECAT scanner at baseline, were performed with GE Discovery 690 (GE Healthcare, Waukesha, WI, US) PET scanner because at that time ECAT EXACT HR + PET scanner was not in use anymore. GE Discovery 690 is a hybrid PET/CT scanner with the axial field of view of 15.7 cm and the patient port is 70.0 cm. Physical performance evaluations of the scanner have shown radial and tangential average spatial resolution of 4.70 mm FWHM and axial resolution of 4.74 mm FWHM [15].

All patients underwent brain MR imaging at baseline and after 3-year follow-up or earlier if the patient was converted to AD. MR imaging was done on Philips 1.5 T Gyroscan Intera CV Nova Dual MR scanner (Philips Medical Systems, Best, The Netherlands). A head coil (Philips Medical Systems, Best, The Netherlands) was used in the measurement. Whole-brain T1-weighted three-dimensional fast field echo data with 1-mm isotropic voxels were acquired in the transverse plane (acquisition parameters: repetition time 25 ms, echo time 5.5 ms, flip angle 30°, field of view 256 × 256 mm) yielding at least 160 contiguous slices to cover the whole head. In addition, routine axial T2-weighted and coronal FLAIR images were obtained.

To obtain quantitative regional values of 18F-Flutemetamol uptake, an automated volume of interest (VOI) analysis was conducted as described previously [16]. Briefly, parametric images representing 18F-Flutemetamol uptake in each pixel were calculated as a region-to-cerebellum ratio of the radioactivity concentration over 90–120 min after 18F-Flutemetamol injection. In order to compensate for head motion during PET imaging, the three ten-minute frames of 18F-Flutemetamol uptake were registered to each other prior to parametric image calculation. A rigid image registration algorithm implemented in statistical parametric mapping (SPM, version 8) software in MATLAB was employed. The motion-corrected data were subsequently summed and co-registered with subject-specific T1-weighted MR images in native space. Finally, the unified segmentation algorithm in SPM8 [17] was used for spatially normalizing the MR and PET images into the standard Montreal Neurological Institute (MNI) coordinate space. Automated VOI-delineation in the MNI space was defined on the basis of automated anatomical labeling [18] atlas, and gray matter masking using individual gray matter segments thresholded at 25%. Standard VOIs of cerebellar gray matter, frontal cortex, parietal cortex, lateral temporal cortex, anterior and posterior cingulate and precuneus were used in the analysis. A composite cortical amyloid uptake score was formed by combining the VOIs of frontal, parietal and lateral temporal cortices and posterior cingulate, similar to our earlier investigations [19]. Average regional VOI-values were extracted from spatially normalized 18F-Flutemetamol uptake ratio images within the above-mentioned VOIs, except in the cerebellar cortex which served as a reference region.

General cognition was assessed using Mini-Mental State Examination (MMSE). A subset of tests defined in Wechsler Memory Scale revised (WMS-R [20]) were conducted to assess immediate recall (logical memory I), delayed recall (logical memory II) and verbal fluency. All tests were delivered yearly to follow-up conversion to AD. Scores from the test occasions closest in time with baseline and follow-up PET imaging were used in the analysis.

VOI-based statistical analysis was conducted using R (version 4.0.3). Descriptive statistics included mean and standard deviations of continuous variables and counts for categorical variables. Student’s t test was employed for paired and Welcher two-sample t test for unpaired analysis. Pearson product moment correlation coefficients were used for examining the relationships between cognitive and imaging measures. Significance level was set at p < 0.05 (trend-level p < 0.1).

Confirmatory voxel-wise analysis was conducted using SPM8. Briefly, the spatially normalized 18F-Flutemetamol uptake ratio images were first smoothed using a kernel of 8 mm (FWHM; 3D) and, secondly, analyzed with paired t test to detect group-level changes in 18F-Flutemetamol uptake ratios over time. The resulting voxel-wise maps of T-statistics were inspected for regional changes of 18F-Flutemetamol uptake ratios using a liberal threshold of T ≥ 2.5, corresponding to approximately p < 0.01 (uncorrected). The use of a liberal threshold allowed detection of clusters that were potentially uncovered using VOI-analysis.