Elsevier

NeuroImage

Volume 52, Issue 4, 1 October 2010, Pages 1367-1373
NeuroImage

Disease modeling in multiple sclerosis: Assessment and quantification of sources of variability in brain parenchymal fraction measurements

https://doi.org/10.1016/j.neuroimage.2010.03.075Get rights and content

Abstract

The measurement of brain atrophy from magnetic resonance imaging (MRI) has become an established method of estimating disease severity and progression in multiple sclerosis (MS). Most commonly reported in the form of brain parenchymal fraction (BPF), it is more sensitive to the degenerative component of the disease and shows progression more reliably than lesion burden. Typically, the reliability of BPF and other morphometric measurements is assessed by evaluating scan–rescan experiments. While these experiments provide good estimates of real-life error related to imperfect patient repositioning in the MRI scanner, measurement variance due to physiological and reversible pathological fluctuations in brain volume are not taken into account.

In this work, we propose a new model for estimating variability in serial morphometry, particularly the BPF measurement. Specifically, we attempt to detect and explicitly model the remaining sources of error to more accurately describe the overall variability in BPF measurements. Our results show that sources of variability beyond subject repositioning error are important and cannot be ignored. We demonstrate that scan–rescan experiments only provide a lower bound on the true error in repeated measurements of patients’ BPF. We have estimated the variance due to patient repositioning during scan–rescan (σsr2 = 3.0e–06), variance assigned to physiological fluctuations (σp2 = 5.74e–06) and the variance associated with lesion activity (σles2 = 1.09e–05). These variance components can be used to determine the relative impact of their sources on sample size estimates for studies investigating change over time in MS patients. Our results demonstrate that sample size calculations based exclusively on scan–rescan variability (σsr) are likely to underestimate the number of patients required. If the physiological variability (σp) is incorporated in sample size calculations, the required sample size would increase by a factor of 5.69 based on standard t-test sample size calculation.

Introduction

Multiple sclerosis (MS) is a progressive degenerative disease in which inflammation, degeneration and repair all contribute to changes in brain volume. Morphometric variables derived from MRI exams are currently the most commonly used structural surrogates of disease severity. The measurement of brain atrophy has become an established method of estimating MS disease severity and progression because it is thought to be more sensitive to the degenerative component of the disease than lesion burden (Bakshi et al., 2008, Bermel and Bakshi, 2006, Zivadinov and Bakshi, 2004, Fisher et al., 2002). Since atrophy is a slow process, a high level of measurement precision is desirable to accurately estimate the true rate of change for a patient (Wei et al., 2004). The most common measure of brain atrophy is brain parenchymal fraction (BPF), but investigations regarding the sources of variability of this measure have been mostly limited to technical aspects (repositioning error, intra- and inter-rater variability of expert driven image analysis components).

A reliable estimate of the precision of morphometric variables is required to draw valid conclusions from longitudinal, follow-up exams. It is also one of the critical parameters in determining minimal follow-up duration or patient sample size for the design of clinical trials and observational studies. The principal sources of variability are:

  • 1.

    Patient re-positioning error

  • 2.

    Algorithm specific errors (e.g., stochastic algorithms or nonlinear algorithm transfer function; intra- and inter-rater variability of expert driven image analysis components)

  • 3.

    Instrument variability (e.g., magnet and/or receiver coil, scanner drift)

  • 4.

    Physiological variations: (caused by events not linked to the disease processes)

  • 5.

    Non-degenerative pathological changes: (e.g., fluctuations in BPF that occur due to disease-related events such as formation of new lesions and associated edema or pseudo-atrophy due to the administration of steroids)

Although there are five sources of variability, the reliability of BPF and other morphometric measurements is frequently measured only with scan–rescan experiments (Wei et al., 2004). In scan–rescan experiments multiple MRI exams are obtained from individual patients within a relatively short interval, under the assumption that no changes due to pathological processes are likely to occur during that time. These experiments provide information only regarding patient repositioning error and algorithm specific errors, and therefore may only provide a lower bound on the true error in repeated measurements of patients.

Other groups have also noted that such scan–rescan experiments approaches may underestimate the sources of variability relevant for longitudinal studies (e.g., subject-related factors, such as hydration status, or instrument-related factors such as field inhomogeneity) (Dickerson et al., 2008, Han et al., 2006). However, there are no quantitative estimates of the extent to which scan–rescan experiments underestimate measurement variability. Also, the effects of subject-related and/or instrument-related factors on sample-size requirements have not been investigated.

In this work, we propose a new model for estimating variability in serial morphometry, particularly the BPF measurement. Specifically, we attempt to detect and explicitly model the remaining sources of error to more accurately describe the overall variability in BPF measurements.

Section snippets

Datasets

In this retrospective study, three datasets were used. The first dataset was from a traditional scan–rescan experiment and consisted of twenty patients diagnosed with clinically definite MS (Guttmann et al., 1999). Brain MR images of each patient were acquired twice within 30 min. All patients were scanned on a 1.5-Tesla machine (GE Signa, General Electrics, Milwaukee, WI) with an axial spin-echo protocol (PDw/T2w, TE 30/80 ms, TR 3000 ms, 192 phase-encoding steps, 256 × 256 × 54 voxels with 0.9375 × 

Results

Table 1 reports summary statistics for the four groups. The mean absolute change in BPF is smallest for group 1 in which the time interval between the two MRI exams was 30 min. Similarly, the lowest variability in the two BPF measurements is seen in group 1. The last column in Table 1 lists the potential sources of variability in the BPF measurements in the various groups.

In Table 2, the coefficient of variation, repeatability coefficient from BA analysis and the SEM and δ values for the four

Discussion

In this work, we propose a new model for estimating variability in serial morphometry, particularly the BPF measurement. Specifically, we attempt to detect and explicitly model the sources of error to more accurately describe the variability in BPF measurements. Using this new model, we have estimated the variance due to patient repositioning during scan–rescan (σsr2), variance ascribed to physiological variations (σp2) and variance ascribed to the presence of enhancing lesions (σles2). Our

Conclusion

We present a new model to estimate the magnitudes of various sources of variability in MRI-derived measurements. We demonstrate that scanrescan experiments may only provide a lower bound on the true error in repeated measurements of patients, and that the variability associated with physiological variation and pathological causes is important and cannot be ignored. We suggest that these observations are important in the design, implementation, analysis, and interpretation of clinical trials.

References (35)

  • C.R. Guttmann et al.

    Quantitative follow-up of patients with multiple sclerosis using MRI: reproducibility

    J. Magn. Reson. Imaging

    (1999)
  • D.S. Marcus et al.

    Open Access Series of Imaging Studies (OASIS): cross-sectional MRI data in young, middle aged, nondemented, and demented older adults

    J. Cogn. Neurosci.

    (2007)
  • X. Wei et al.

    Quantitative analysis of MRI signal abnormalities of brain white matter with high reproducibility and accuracy

    J. Magn. Reson. Imaging

    (2002)
  • A.F. Fotenos et al.

    Normative estimates of cross-sectional and longitudinal brain volume decline in aging and AD

    Neurology

    (2005)
  • Y. Zhang et al.

    Segmentation of brain MR images through a hidden Markov random field model and the expectation-maximization algorithm

    IEEE Trans Med Imaging

    (2001)
  • M. Filippi et al.

    Intraobserver and interobserver variability in schemes for estimating volume of brain lesions on MR images in multiple sclerosis

    AJNR Am. J. Neuroradiol.

    (1998)
  • J.M. Bland et al.

    Statistical methods for assessing agreement between two methods of clinical measurement

    Lancet

    (1986)
  • Cited by (24)

    • Brain volume loss in individuals over time: Source of variance and limits of detectability

      2020, NeuroImage
      Citation Excerpt :

      However, the fluctuations in measured brain volume seen on repeated scans of individuals can be much greater than the mean magnitude of brain volume fluctuations seen on scan-reposition-rescan experiments for a group (Biberacher et al., 2016), and can be on the same order as the change in brain volume expected in MS patients over one year. Since changes of this magnitude can be induced over the course of a few hours in dehydration/rehydration experiments (Duning et al., 2005; Kempton et al., 2009; Nakamura et al., 2014a), it is plausible that normal physiological fluctuations in brain water content over the course of a day (e.g. before vs. after meals, before or after coffee consumption, morning vs. evening) could dominate short-term fluctuations (on the time scale of hours) of brain volume in individuals (Sampat et al., 2010; Nakamura et al., 2015). In addition, disease-related processes other than irreversible tissue loss, such as development or resolution of focal or diffuse inflammation in a person with MS, could contribute to changes in brain volume in individuals on the time scale of days to months (Sampat et al., 2010; De Stefano and Arnold, 2015; Rudick et al., 1999).

    • Novel multi-linear quantitative brain volume formula for manual radiological evaluation of brain atrophy

      2020, European Journal of Radiology Open
      Citation Excerpt :

      The assumption is that the skull volume is the normal standard reference of each individual and the deviation of brain volume from its edges is the measure of atrophic change since the ratio between brain volume and cranial cavity is constant in early age [24]. This simple reproducible formula is a linear synthesis of the popularly known Brain Parenchymal Fraction (BPF) which is automated quantitative method [25]. Inasmuch as the manual evaluation methods remain to be alternative techniques when automated methods cannot be accessed, this study was done to be part of the solution to such limitations.

    • Intra- and interscanner variability of magnetic resonance imaging based volumetry in multiple sclerosis

      2016, NeuroImage
      Citation Excerpt :

      Despite the broad practice of multicenter and longitudinal MRI studies in MS, intrascanner in relation to interscanner variability has not yet been investigated in MS patients. In addition to the sources of variance listed above, disease-related effects beyond the object of investigation have to be considered: MS lesions and diffuse brain tissue changes disturb automated image analysis (Chard et al., 2002) and add additional variance (Sampat et al., 2010). The disease might cause partially opposing effects on measurements with inflammation leading to volume increase as well as neurodegeneration and demyelination leading to volume decrease.

    View all citing articles on Scopus
    View full text