Email updates

Keep up to date with the latest news and content from BMC Medical Imaging and BioMed Central.

Open Access Technical advance

Improved assessment of multiple sclerosis lesion segmentation agreement via detection and outline error estimates

David S Wack125*, Michael G Dwyer1, Niels Bergsland1, Carol Di Perri3, Laura Ranza3, Sara Hussein1, Deepa Ramasamy1, Guy Poloni1 and Robert Zivadinov14

Author Affiliations

1 Buffalo Neuroimaging Analysis Center, Dept. of Neurology, University at Buffalo, State University of New York at Buffalo, Buffalo, NY, USA

2 Department of Nuclear Medicine, University at Buffalo, State University of New York at Buffalo, Buffalo, NY, USA

3 Department of Neuroradiology, IRCCS, C. Mondino, University of Pavia, Pavia, Italy

4 The Jacobs Neurological Institute, Dept. of Neurology, University at Buffalo, State University of New York at Buffalo, Buffalo, NY, USA

5 Buffalo Neuroimaging Analysis Center, Jacobs Neurological Institute, State University of NY at Buffalo, 100 High St., Buffalo, NY, 14203, USA

For all author emails, please log on.

BMC Medical Imaging 2012, 12:17  doi:10.1186/1471-2342-12-17

Published: 19 July 2012

Abstract

Background

Presented is the method “Detection and Outline Error Estimates” (DOEE) for assessing rater agreement in the delineation of multiple sclerosis (MS) lesions. The DOEE method divides operator or rater assessment into two parts: 1) Detection Error (DE) -- rater agreement in detecting the same regions to mark, and 2) Outline Error (OE) -- agreement of the raters in outlining of the same lesion.

Methods

DE, OE and Similarity Index (SI) values were calculated for two raters tested on a set of 17 fluid-attenuated inversion-recovery (FLAIR) images of patients with MS. DE, OE, and SI values were tested for dependence with mean total area (MTA) of the raters' Region of Interests (ROIs).

Results

When correlated with MTA, neither DE (ρ = .056, p=.83) nor the ratio of OE to MTA (ρ = .23, p=.37), referred to as Outline Error Rate (OER), exhibited significant correlation. In contrast, SI is found to be strongly correlated with MTA (ρ = .75, p < .001). Furthermore, DE and OER values can be used to model the variation in SI with MTA.

Conclusions

The DE and OER indices are proposed as a better method than SI for comparing rater agreement of ROIs, which also provide specific information for raters to improve their agreement.

Keywords:
Multiple sclerosis; Detection and outline error estimates; Rater agreement; Operator agreement; Metric; Jaccard Index; Similarity index; Measure; Index; Kappa; Lesion; MRI; ROI