A comparison of transverse relaxation time estimation models for cardiac magnetic resonance imaging in assessing left ventricular iron overload

Cardiac magnetic resonance imaging is one of the most advanced tools used in clinical routine to assess the physiology, anatomy or function of the heart. Its versatility and non-invasiveness have made it an important method of diagnosis. This technique can also be exploited to measure myocardial iron overload, since iron is a paramagnetic substance. Iron accumulation leads to a decrease in T2* relaxation time, which can be used as a marker to detect important diseases such as Haemochromatosis and Thalassaemia. A significant reduction in T2* means that there is a substantial accumulation of iron, resulting in myocardial dysfunction. To assess left ventricular iron overload, four T2* estimation models were implemented: three non-linear models and one linear model, comparing pixel-wise approach and ROI-based approach. Since the dependence of the MR signal on echo times can be described by an exponential decay, the implemented non-linear models are distinguished between a single-exponential model, a biexponential model and an offset model. The fourth model was obtained by linearising the exponential model. From each model, it is thus possible to obtain T2* maps of the ventricular myocardium, which is divided into three parts: basal, medial and apical region, for a total of sixteen segments. All these models were compared using different evaluation metrics, and mono-exponential model proved to be the most suitable for T2* estimation. Besides, all models were applied to healthy subjects and patients with Haemochromatosis, to evaluate T2* trend in all segments. Using T2* estimates obtained in healthy subjects, a map of correction factors was also created to correct for the presence of artifacts such as the heart-lung interface and blood vessel pulsation that corrupt the image signal. Even if the calculation of the correction factors is based on manual segmentations, thus involving a selection of the mid-ventricular septum by the radiologist, the map obtained turns out to be robust also to simulated error of the individuation of the septum. In addition, a second correction map was implemented based on more precise segmentations, which do not divide the myocardium into equidistant segments, but adjust the segment lengths based on the inter-ventricular septum. In conclusion, T2* parametric mapping proves to be a good assessment method to reveal the presence of myocardial iron overload, allowing accurate diagnoses.