Meeting Banner
Abstract #3943

Myocardial T2* Cut-Off Values Determined by Gaussian Mixtures Models

Taigang He1, 2, Yanqiu Feng3, Rongrong Liu4, John-Paul Carpenter1, Andrew Jabbour1, Yongrong lai4, Dudley Pennell1, David Firmin1, 2

1CMR Unit, Royal Brompton Hospital, London, United Kingdom; 2NHLI, Imperial College, London, United Kingdom; 3School of Biomedical Engineering, Sourthern Medical University, China; 4Department of Hematology, The first affiliated Hospital of Guangxi Medical University, China

Cardiovascular magnetic resonance (CMR) T2* can provide a non-invasive means of measuring tissue iron in the heart. We have found that T2*< 20ms represents iron overload in the heart, and that T2*<10ms predicts risk of heart faliure in patients with thalassemia major. In this study, we aimed at clustering patients automatically into normal and abnormal groups by using Gaussian mixture models. This study confirms statistically that T2* is a reliable tool for screening patients with iron overload. The results agree well with T2* cut-off value of 20ms and 10ms commonly used in clinical practice.

Keywords

abnormal accepted advanced affiliated agree agrees aimed another array assessment automated automatically axis believe better biomedical bishop black blood breath cardiac cardiovascular carpenter characterization china circulation clarify clinical cluster clustered clustering coil college commonly complementary confirm confirms consistent contain corresponds currently curve cutoff decay deposition determined developed development diagnosis dimensional distinctive distinguish distribution divided early employed engineering exists explain explored exponential failure finding findings fitting future gating good green heart hematology histogram hold hospital imperial important improved indicate indicates intrinsic invasive investigation iron john kingdom kirk learning limited machine major males material mature measured measuring medical mismatched mixture mixtures model models mono myocardial needed nonlinear observations overall overload patient patients pattern patterns people population practice predict predicts previous probability proton proven recognition reliable represents republic risk royal samples scanned scanner school screening septum short shortening slice sonata springer statistically suggests technical tissue tool unit useful validated ventricular widely years