[18F]AV45 standardized uptake value ratio harmonization using ComBat in multi‐center cross‐sectional Alzheimer’s studies
Tahnia Nazneen, Stijn Servaes, Seyyed Ali Hosseini, Cécile Tissot, Nesrine Rahmouni, Joseph Therriault, Arthur C. Macedo, Dana Tudorascu, Pedro Rosa‐Neto- Psychiatry and Mental health
- Cellular and Molecular Neuroscience
- Geriatrics and Gerontology
- Neurology (clinical)
- Developmental Neuroscience
- Health Policy
- Epidemiology
Abstract
Background
[18F]AV45 (florbetapir) based Positron emission tomography (PET) imaging of brain amyloid load is one of the core biomarkers for Alzheimer’s disease (AD). However, with the rise in the trend of combining data from multicenter studies, the need to correct substantial technical variability associated with image intensity scale due to multi‐center effect continues to rise. While smoothing methods remove small variability, data‐driven feature harmonization methods can adjust for scanner settings and patient motion without over‐blurring the scans. Here we aimed to investigate the effect of ComBat harmonization on multi‐site florbetapir studies to reduce variance across diagnosis groups.
Method
We assessed 163 scans from the Alzheimer’s disease Neuroimaging Initiative (ADNI3) database. These scans were acquired from 14 different sites. T1‐weighted MRI images were processed using the ADNI pipeline. PET images had 20 min (4×5min frames) acquisition at 50‐70 min post‐injection of 370 MBq (10.0 mCi) ± 10% florbetapir. Raw PET images from all sites were downloaded for quality control at the University of Michigan where the rest of the preprocessing took place using ADNI guidelines. These preprocessed scans were then used to extract SUVR maps using the cerebellar gray matter as the reference region. Based on the literature, the global SUVR for each subject was estimated from the averaged frontal, parietal, temporal, and cingulate cortices. ComBat harmonization was then performed using multicenter data, preserving diagnosis group as the covariate. A paired t‐test and logistic regression were performed to analyze the effect of the harmonization method on the SUVR.
Result
A paired t‐test confirms the significant difference between the pre‐and post‐ComBat SUVR, especially for AD samples. Although the variance only decreases by 7.1% for CN as opposed to 26.9% (MCI) and 55.4% (AD), this may be explained by the large variance in the age in the sample as well as the possibility of amyloid positive individuals in the CN population. The logistic regression shows an accuracy of 87.9% for AD.
Conclusion
Even though ComBat harmonization significantly reduces the variance in AD and MCI population, more studies need to be performed to check its use across all PET derived features.