Reproducibility and repeatability of magnetic resonance imaging in dementia

Highlights

• Reproducibility and repeatability of a dementia MRI protocol were assessed.

• Functional imaging markers were more variable than structural ones.

• Short-term repeatability was better than long-term repeatability.

• Reproducibility was comparable to long-term repeatability.

• Results support viability of multi-site, longitudinal studies of cognitive decline.

Abstract

Purpose

Individualised predictive models of cognitive decline require disease-monitoring markers that are repeatable. For wide-spread adoption, such markers also need to be reproducible at different locations. This study assessed the repeatability and reproducibility of MRI markers derived from a dementia protocol.

Methods

Six participants were scanned at three different sites with a 3T MRI scanner. The protocol employed: T1-weighted (T1w) imaging, resting state functional MRI (rsfMRI), arterial spin labelling (ASL), diffusion-weighted imaging (DWI), T2-weighted fluid attenuation inversion recovery (FLAIR), T2-weighted (T2w) imaging, and susceptibility weighted imaging (SWI). Participants were scanned repeatedly, up to six times over a maximum period of five years. One participant was also scanned a further three times on sequential days on one scanner. Fifteen derived metrics were computed from the seven different modalities.

Results

Reproducibility (coefficient of variation; CoV, across sites) was best for T1w derived grey matter, white matter and hippocampal volume (CoV < 1.5%), compared to rsfMRI and SWI derived metrics (CoV, 19% and 21%). For a given metric, long-term repeatability (CoV across time) was comparable to reproducibility, with short-term repeatability considerably better.

Conclusions

Reproducibility and repeatability were assessed for a suite of markers calculated from a dementia MRI protocol. In general, structural markers were less variable than functional MRI markers. Variability over time on the same scanner was comparable to variability measured across different scanners. Overall, the results support the viability of multi-site longitudinal studies for monitoring cognitive decline.

Introduction

People with mild cognitive impairment (MCI) [1] show age-related decline that is greater than that of their age-matched peers. A diagnosis of dementia is given when a significant loss of everyday cognitive function, unrelated to frailty, is identified. Approximately 50 million people are living with dementia worldwide, a number set to increase three-fold by 2050 [2]. Although MCI can be a precursor, not all people with MCI go on to develop dementia [3]. The reasons why some individuals progress to dementia and others do not is unresolved and remains a topic of focussed research. Biomarkers are needed that accurately track cognitive decline and hold potential to predict which individuals will progress to dementia.

Quantitative imaging biomarkers (QIBs) assist with diagnosis, disease-monitoring, and assessment of treatment and interventions. QIBs derived from magnetic resonance imaging (MRI) that are of interest in dementia studies [4] include grey matter brain volumes derived from T1-weighted (T1w) imaging for tissue atrophy, arterial spin labelling (ASL) metrics for cerebral blood flow (hypo-perfusion/hypo-metabolism), diffusion weighted imaging (DWI) measures for white matter structure, resting state functional MRI (rsfMRI) for functional connectivity and T2w fluid attenuation inversion recovery (FLAIR) derived estimates of white matter hyperintensity (WMH) volume for small vessel disease. Additional, more clinically focused scans in a dementia protocol often include T2-weighted (T2w) and susceptibility weighted imaging (SWI) to assess vascular health and other pathologies.

Multicentre studies allow for larger, more representative cohorts to be recruited for research trials. However, inter-site measurement variability needs to be quantifiable to interpret pooled data. For QIBs to be adopted widely and incorporated into clinical practice, inter-site measurement reliability needs to be established to determine confidence in diagnostic metrics. Knowledge of QIB variability over time due to measurement uncertainty is essential when monitoring longitudinal cognitive changes as a function of normal ageing, disease, or an intervention. Without this crucial information, it cannot be determined whether subsequent changes in the QIB are due to underlying physiological changes, or simply measurement variability.

The Quantitative Imaging Biomarkers Alliance (QIBA) [5], [6] defines reproducibility as variability due to measurements being collected under different conditions, e.g., at different sites with different hardware or processed with different software. Conversely repeatability refers to variability in a measurement collected under the same conditions multiple times. Several studies have examined the reproducibility and repeatability of individual, or single modality QIBs within the context of dementia [7], [8], [9]. However multiple QIBs in combination, akin to a “biomarker signature”, are likely to have more predictive power of cognitive decline than a single maker alone [10]. Accordingly, we wished to assess measurement variability in a suite of parameters that could be used for this purpose. We recruited “travelling heads” (THs), the same participants who travelled to imaging centres to be scanned at repeated time-points, enabling assessment of the reproducibility and repeatability of quantitative MRI (qMRI) markers for dementia. Each site was part of a multicentre, longitudinal study known as the Dementia Prevention Research Clinics.