Improving the spatial resolution of Cardiac Magnetic Resonance Quantitative T1-mapping using a super-resolution reconstruction
To treat atrial fibrillation, many patients require radiofrequency ablation (RFA), for which myocardial fibrosis is a predictor of success. T1 mapping is currently used in the clinic for tissue characterization (and fibrosis identification) in the ventricles, but its application to the left atrium remains challenging due to its smaller dimensions. To ensure that a sufficiently high spatial resolution is attained within acquisition times compatible with a clinical protocol, strategies for acquisition acceleration are required.
The aim of this project is to implement a reconstruction pipeline for performing super-resolution in the context of T1-mapping, exploring the robustness of different k-space sampling patterns to respiratory and cardiac motion.
Comparison of T1-maps and Late Gadolinium Enhancement in the detection of myocardial fibrosis
In many heart muscle pathologies, the characterization of myocardium fibrosis is a critical step for understanding disease progression and guide patient treatment. The gold standard technique for non-invasive imaging of this tissue is magnetic resonance imaging (MRI) with late gadolinium enhancement (LGE). In alternative, T1 mapping techniques have been shown to be advantageous compared to LGE as they allow an accurate, reproducible, and quantitative assessment of fibrosis without the need for contrast agent injection. However, the exact relationship between the tissue’ changes identified in LGE and T1-maps still remains under research.
The aim of this project is to investigate the value of T1 maps in the identification of fibrosis by comparison with the standard LGE images when using machine learning analysis. These approaches will be applied to previously collected datasets, which have already been qualitatively evaluated by our clinical collaborators at Hospital da Luz.
Optimizing T2 quantitative estimation of knee cartilage with Magnetic Resonance Imaging (MRI) applying dictionary-based methods
Cartilage osteoarthritis causes severe disability, affecting 70 to 90% of patients over 65 years old. Knee osteoarthritis is commonly assessed with X-ray imaging. However, research studies suggest that cartilage degeneration is visible in those images, only when its function and mobility have already been compromised. Recent studies suggest T2 and T1p quantitative MRI may be adequate predictors of early osteoarthritis and could potentially provide valuable information on disease progression, useful for patient follow-up and therapeutic decision.
The aim of this project is to implement an estimation algorithm for T2 mapping where exact knowledge of the used multi spin-echo sequence – MSE is used to predict the observed signal and finding the best match out of a range of plausible T2 values, and apply it to the knee cartilage (Figure shows results from a pilot test). Resulting T2 maps of the knee will then be compared to those obtained with the gold-standard estimation methods (e.g. mono-exponential fit, disregarding the first echo).
Figure: MSE T2-weighted images of the knee joint with superimposed T2 map estimation of the selected knee cartilage. Comparison of the gold-standard method using a mono-exponential decay model (left) and the proposed dictionary-based method (right). Slight differences can be noticed in the T2 cartilage estimated values (colour scaling represents T2 times in ms per pixel). Data was acquired at Hospital de Santo António, in collaboration with José Manuel Coelho, António Oliveira and Luísa Nogueira.
Comparing free-water fraction estimation algorithms in the context of Traumatic Brain Injury
Advisership – IST: Rita Nunes
University of Cambridge: Marta Correia
Traumatic Brain Injury (TBI) is defined as an alteration of brain function or other evidence of brain pathology, caused by an external force. Worldwide, TBI is a leading cause of injury-related death and disability, with a devastating impact on patients and their families. Diffusion Weighted MRI (DWI) can be used to explore the microstructural architecture of brain tissues in vivo and it has been widely applied to the study of TBI pathology. However, the presence of extracellular free water from edema can affect the diffusion measures, potentially leading to wrong interpretations about the underlying microstructural changes. The free-water elimination (FWE) signal model is an alternative to more traditional approaches to modelling of DWI data, which considers also an isotropic extracellular compartment representing free water. A recent method for estimating free water fraction using multiple diffusion-weighting shells has been shown to reduce the bias in the parameter estimates. However, as clinical protocols often use a single diffusion-weighing (single-shell data) to reduce exam times, it becomes relevant to investigate if introducing prior knowledge in the estimation could enable reliable free water elimination when applied to single-shell data.
The goal of this project is to compare the performance of these two free water elimination algorithms when applied to the same TBI data. A large dataset of multi-shell DWI data has been acquired as part of a longitudinal TBI study led by the NTNU in Trondheim. This dataset includes 155 patients and 78 healthy controls scanned at multiple time points (622 scans in total). The data has already been pre-processed, and the multi-shell algorithm applied to eliminate free water contamination. In this project the same data will be processed applying an open-source implementation of the single-shell algorithm. The results will be compared to the output of the multi-shell algorithm (currently the state-of-the art in the field) for the same data, in order to assess the reliability of the diffusion metrics estimated by the single-shell approach.