POSTER PRESENTATION Open Access Detection and correction ... · POSTER PRESENTATION Open Access...
2
POSTER PRESENTATION Open Access Detection and correction of regional shape bias arising from imaging protocol: differences between GRE and SSFP Pau Medrano-Gracia 1* , David A Bluemke 2 , Brett R Cowan 1 , J Paul Finn 3 , Daniel C Lee 5 , Joao A Lima 4 , Avan Suinesiaputra 1 , Alistair A Young 1 From 15th Annual SCMR Scientific Sessions Orlando, FL, USA. 2-5 February 2012 Summary We present a methodology to detect and correct shape bias arising from imaging protocol in an atlas of the left ventricle. We show how it can be used to correct the differences between GRE and SSFP shapes and volumes. Background It is well established that different imaging protocols lead to different clinical parameters, for example gradi- ent recalled echo (GRE) and steady state free precession (SSFP) protocols lead to different cardiac mass and volumes[1]. However, it is not known whether these dif- ferences are local or global in nature. Correction of local biases in heart shape is required for patient follow- up and meta-analyses of clinical trials which acquire data using different protocols. For example, the Cardiac Atlas Project (CAP)[2] includes GRE images of asymptomatic volunteers from the MESA study[3]and patients with myocardial infarc- tion imaged by SSFP from the DETERMINE study [4]. By computing a shape mapping from one protocol to another, we can establish a bias-free comparison between these two populations. Methods Finite element shape models of the left ventricle (LV) were customized to 46 normal volunteers imaged by both GRE and SSFP and a multi-dimensional statistical transform was developed to correct the systematic bias between the protocols. Given that in our method there is no a priori guarantee that the mapping of local parameters will also obey the previously reported differ- ences in terms of mass and volumes, errors were exam- ined using both local shape and global volumetric measures. Results There was a systematic local difference between GRE and SSFP around the apex and papillary muscles at ED, and also around the base at ES (Fig. 1) which is prob- ably due to the flow induced variation in blood-myocar- dial contrast in GRE imaging. The bias was corrected locally and the mapping also provided a global correc- tion for volume (Table 1). Further, cross-validation experiments showed that this methodology only required ~25 cases with both modalities to determine the transformation to robustly convert between scanning techniques. This methodology can also be successfully applied to other types of bias, for instance, that of different analysts or, between data processed by different laboratories. Conclusions GRE and SSFP cardiac imaging protocols give rise to regional differences in LV geometry. We have estab- lished a generalisable framework which removes these biases and also corrects for mass and volumes. Funding NIH - National Heart, Lung and Blood Institute (Award Number R01HL087773). Author details 1 Auckland Bioengineering Institute, The University of Auckland, Auckland, New Zealand. 2 National Institute of Biomedical Imaging and Bioengineering, 1 Auckland Bioengineering Institute, The University of Auckland, Auckland, New Zealand Full list of author information is available at the end of the article Medrano-Gracia et al. Journal of Cardiovascular Magnetic Resonance 2012, 14(Suppl 1):P270 http://www.jcmr-online.com/content/14/S1/P270 © 2012 Medrano-Gracia et al; licensee BioMed Central Ltd. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
Transcript of POSTER PRESENTATION Open Access Detection and correction ... · POSTER PRESENTATION Open Access...
POSTER PRESENTATION Open Access
Detection and correction of regional shape biasarising from imaging protocol: differencesbetween GRE and SSFPPau Medrano-Gracia1*, David A Bluemke2, Brett R Cowan1, J Paul Finn3, Daniel C Lee5, Joao A Lima4,Avan Suinesiaputra1, Alistair A Young1
From 15th Annual SCMR Scientific SessionsOrlando, FL, USA. 2-5 February 2012
SummaryWe present a methodology to detect and correct shapebias arising from imaging protocol in an atlas of the leftventricle. We show how it can be used to correct thedifferences between GRE and SSFP shapes and volumes.
BackgroundIt is well established that different imaging protocolslead to different clinical parameters, for example gradi-ent recalled echo (GRE) and steady state free precession(SSFP) protocols lead to different cardiac mass andvolumes[1]. However, it is not known whether these dif-ferences are local or global in nature. Correction oflocal biases in heart shape is required for patient follow-up and meta-analyses of clinical trials which acquiredata using different protocols.For example, the Cardiac Atlas Project (CAP)[2]
includes GRE images of asymptomatic volunteers fromthe MESA study[3]and patients with myocardial infarc-tion imaged by SSFP from the DETERMINE study [4].By computing a shape mapping from one protocol toanother, we can establish a bias-free comparisonbetween these two populations.
MethodsFinite element shape models of the left ventricle (LV)were customized to 46 normal volunteers imaged byboth GRE and SSFP and a multi-dimensional statisticaltransform was developed to correct the systematic biasbetween the protocols. Given that in our method thereis no a priori guarantee that the mapping of local
parameters will also obey the previously reported differ-ences in terms of mass and volumes, errors were exam-ined using both local shape and global volumetricmeasures.
ResultsThere was a systematic local difference between GREand SSFP around the apex and papillary muscles at ED,and also around the base at ES (Fig. 1) which is prob-ably due to the flow induced variation in blood-myocar-dial contrast in GRE imaging. The bias was correctedlocally and the mapping also provided a global correc-tion for volume (Table 1). Further, cross-validationexperiments showed that this methodology onlyrequired ~25 cases with both modalities to determinethe transformation to robustly convert between scanningtechniques.This methodology can also be successfully applied to
other types of bias, for instance, that of different analystsor, between data processed by different laboratories.
ConclusionsGRE and SSFP cardiac imaging protocols give rise toregional differences in LV geometry. We have estab-lished a generalisable framework which removes thesebiases and also corrects for mass and volumes.
FundingNIH - National Heart, Lung and Blood Institute (AwardNumber R01HL087773).
Author details1Auckland Bioengineering Institute, The University of Auckland, Auckland,New Zealand. 2National Institute of Biomedical Imaging and Bioengineering,
1Auckland Bioengineering Institute, The University of Auckland, Auckland,New ZealandFull list of author information is available at the end of the article
Medrano-Gracia et al. Journal of Cardiovascular Magnetic Resonance 2012, 14(Suppl 1):P270http://www.jcmr-online.com/content/14/S1/P270
© 2012 Medrano-Gracia et al; licensee BioMed Central Ltd. This is an open access article distributed under the terms of the CreativeCommons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, andreproduction in any medium, provided the original work is properly cited.
NIH Clinical Center, Bethesda, MD, USA. 3Diagnostic CardioVascular Imaging,UCLA, Los Angeles, CA, USA. 4D. W. R. Cardiovascular Clinical ResearchCenter, Johns Hopkins University, Baltimore, MD, USA. 5Division ofCardiology, Northwestern University, Chicago, IL, USA.
Published: 1 February 2012
References1. Malayeri AA, et al: Journal of Magnetic Resonance Imaging. 2008, 28:60.2. Fonseca , Carissa G, et al: Bioinformatics (Oxford Journals). 2011, doi:
10.1093/bioinformatics/btr360.3. Bild DE, et al: American Journal of Epidemiology. 2002, 156:871.4. Kadish AH, et al: Journal of Cardiovascular Electrophysiology. 2009,
20:982-987.
doi:10.1186/1532-429X-14-S1-P270Cite this article as: Medrano-Gracia et al.: Detection and correction ofregional shape bias arising from imaging protocol: differences betweenGRE and SSFP. Journal of Cardiovascular Magnetic Resonance 2012 14(Suppl 1):P270.
Submit your next manuscript to BioMed Centraland take full advantage of:
• Convenient online submission
• Thorough peer review
• No space constraints or color figure charges
• Immediate publication on acceptance
• Inclusion in PubMed, CAS, Scopus and Google Scholar
• Research which is freely available for redistribution
Submit your manuscript at www.biomedcentral.com/submit
Figure 1 Shape bias between GRE and SSFP at ED (left) and ES (right); the arrow points from the centre toward the septum.
Table 1 Volume comparison of the original volumesbetween GRE and SSFP and the estimated ones from theleave-one-out experiments (L1). The error in volume iscomputed using the SSFP volume as reference.
Cavity vol. ED (ml) GRE SSFP Estimated SSFP (L1)
Mean (μ) 126.2 134.8 135.6
Std. dev. (s) 27.0 28.4 29.0
Error mean -8.6 – 0.9
Error std. dev. 12.4 – 12.9
Cavity vol. ES (ml)
Mean (μ) 52.8 53.5 54.3
Std. dev. (s) 12.6 13.8 13.4
Error mean -0.6 – 0.9
Error std. dev. 8.2 – 8.8
LV mass ED (g)
Mean (μ) 145.3 132.3 130.9
Std. dev. (s) 33.1 32.1 30.7
Error mean 12.34 – -1.2
Error std. dev. 10.16 – 11.1
Medrano-Gracia et al. Journal of Cardiovascular Magnetic Resonance 2012, 14(Suppl 1):P270http://www.jcmr-online.com/content/14/S1/P270
Page 2 of 2
