Contributions
Abstract: P520
Type: Poster
Abstract Category: Pathology and pathogenesis of MS - 21 Imaging
Background and purpose: Sharing magnetic resonance (MR) images and other information from multiple sclerosis (MS) patients offers many benefits to MS research. However, in this context protecting patients' privacy is crucial. In addition to removing identifying metadata, MR images should be defaced (i.e. facial features removed from images). Yet, it is unknown whether using defaced images affects assessment of important features from MR in MS. For three recent face removal methods, we assessed their impact on automated MS lesion and whole brain volume measurement.
Material and method: 100 MS patients all with 3.0 T 3D FLAIR and 3D T1 images were included. Images were defaced using Quickshear (Schimke; 2011), Facemasking (Milchenko; 2012), and Defacing (Bischoff-Grethe; 2007). On native images and each type of defaced images, lesion volumes were quantified using LST (www.statistical-modelling.de/lst.html) and whole brain volumes were quantified using SIENAX (Smith; 2002). We first assessed whether defacing introduced systematic changes of whole brain or lesions volumes, by using repeated measures ANOVA with defacing method as categorical variable and testing its effect by an F-test. Next, to evaluate volumetric agreement, we quantified the intraclass correlation coefficient (ICC) for absolute agreement.
Results: At the group level, whole brain and lesion volumes were highly similar between native and defaced images. Median [interquartile range, IQR] lesion volumes were 15.37 [5.67-30.86] (native), 15.46 [5.75-31.96] (Quickshear), 16.42 [6.18-31.96] (Facemasking) and 15.53 [5.87-31.17] mL (Defacing). Median[IQR] normalized whole brain volumes were 1.45 [1.38-1.50] L (native), 1.45 [1.39-1.51] L (Quickshear), 1.46 [1.40-1.52] L (Facemasking) and 1.45 [1.39-1.51] L (Defacing). The defacing methods had no systematic effects on lesion volumes (p=0.46) or whole brain volumes (p=0.99) (F-tests). However, defaced volumes did differ from native volumes: For lesion volumes, ICC with native images was poor for Defacing (ICC=0.23), fair for Quickshear (0.54), and good for Facemasking (0.61). For whole brain volume, ICC was fair for Quickshear (0.59) and Defacing (0.58), and good for Facemasking (0.67).
Conclusion: Defacing has severe impact on the automated assessment of MS lesion volumes and brain volumes by currently popular software methods. Data sharing initiatives in MS research should devise methods to resolve these current shortcomings.
Disclosure:
A. de Sitter is employed on a project sponsored by a research grant from Teva Pharmaceuticals (grant to H. Vrenken and F. Barkhof).
M. Visser nothing to disclose.
I. Brouwer is partly employed on projects sponsored by research grants from Teva Pharmaceuticals and Novartis Pharma (grants to H. Vrenken and F. Barkhof).
R.A. van Schijndel is partly working for the Image Analysis Center, a contract research organization of the VU University Medical Center.
B.M.J. Uitdehaag has received personal compensation for consulting from Biogen Idec, Genzyme, Merck Serono, Novartis, Roche, and TEVA.
F. Barkhof has received compensation for consulting services and/or speaking activities from Bayer Schering Pharma, Biogen Idec, Merck Serono, Novartis, Genzyme, Synthon BV, Roche, Teva, Jansen research and IXICO.
H. Vrenken has received research grants from Pfizer, MerckSerono, Novartis and Teva, speaker honoraria from Novartis, and consulting fees from MerckSerono; all funds were paid directly to his institution.
Abstract: P520
Type: Poster
Abstract Category: Pathology and pathogenesis of MS - 21 Imaging
Background and purpose: Sharing magnetic resonance (MR) images and other information from multiple sclerosis (MS) patients offers many benefits to MS research. However, in this context protecting patients' privacy is crucial. In addition to removing identifying metadata, MR images should be defaced (i.e. facial features removed from images). Yet, it is unknown whether using defaced images affects assessment of important features from MR in MS. For three recent face removal methods, we assessed their impact on automated MS lesion and whole brain volume measurement.
Material and method: 100 MS patients all with 3.0 T 3D FLAIR and 3D T1 images were included. Images were defaced using Quickshear (Schimke; 2011), Facemasking (Milchenko; 2012), and Defacing (Bischoff-Grethe; 2007). On native images and each type of defaced images, lesion volumes were quantified using LST (www.statistical-modelling.de/lst.html) and whole brain volumes were quantified using SIENAX (Smith; 2002). We first assessed whether defacing introduced systematic changes of whole brain or lesions volumes, by using repeated measures ANOVA with defacing method as categorical variable and testing its effect by an F-test. Next, to evaluate volumetric agreement, we quantified the intraclass correlation coefficient (ICC) for absolute agreement.
Results: At the group level, whole brain and lesion volumes were highly similar between native and defaced images. Median [interquartile range, IQR] lesion volumes were 15.37 [5.67-30.86] (native), 15.46 [5.75-31.96] (Quickshear), 16.42 [6.18-31.96] (Facemasking) and 15.53 [5.87-31.17] mL (Defacing). Median[IQR] normalized whole brain volumes were 1.45 [1.38-1.50] L (native), 1.45 [1.39-1.51] L (Quickshear), 1.46 [1.40-1.52] L (Facemasking) and 1.45 [1.39-1.51] L (Defacing). The defacing methods had no systematic effects on lesion volumes (p=0.46) or whole brain volumes (p=0.99) (F-tests). However, defaced volumes did differ from native volumes: For lesion volumes, ICC with native images was poor for Defacing (ICC=0.23), fair for Quickshear (0.54), and good for Facemasking (0.61). For whole brain volume, ICC was fair for Quickshear (0.59) and Defacing (0.58), and good for Facemasking (0.67).
Conclusion: Defacing has severe impact on the automated assessment of MS lesion volumes and brain volumes by currently popular software methods. Data sharing initiatives in MS research should devise methods to resolve these current shortcomings.
Disclosure:
A. de Sitter is employed on a project sponsored by a research grant from Teva Pharmaceuticals (grant to H. Vrenken and F. Barkhof).
M. Visser nothing to disclose.
I. Brouwer is partly employed on projects sponsored by research grants from Teva Pharmaceuticals and Novartis Pharma (grants to H. Vrenken and F. Barkhof).
R.A. van Schijndel is partly working for the Image Analysis Center, a contract research organization of the VU University Medical Center.
B.M.J. Uitdehaag has received personal compensation for consulting from Biogen Idec, Genzyme, Merck Serono, Novartis, Roche, and TEVA.
F. Barkhof has received compensation for consulting services and/or speaking activities from Bayer Schering Pharma, Biogen Idec, Merck Serono, Novartis, Genzyme, Synthon BV, Roche, Teva, Jansen research and IXICO.
H. Vrenken has received research grants from Pfizer, MerckSerono, Novartis and Teva, speaker honoraria from Novartis, and consulting fees from MerckSerono; all funds were paid directly to his institution.