Oportunidades Públicas

Reconstruction of incomplete images

The images submuestreadas are a way to increase the speed of acquisition for the magnetic resonance imaging. In this project we will study different spaces in which they can represent the data in a manner to facilitate its reconstruction by employing algorithms of Compressed Sensing.

Magnetic Resonance Imaging Acquisition and Reconstruction

Deep learning based super resolution reconstruction for high resolution coronary magnetic resonance angiography

Coronary artery disease is currently diagnosed with X-ray or computed tomography coronary angiography (CTCA). However, both modalities expose patients to ionising radiation and iodinated contrast agents (CA). We have developed an alternative approach, coronary magnetic resonance angiography (CMRA), which is non-invasive, does not require the injection of CA and is free of radiation. Our 3D CMRA technique is approaching the resolution of CTCA but still requires a 10-minute scan. To further shorten the scan time to <1 minute and match the resolution of CTCA (0.6mm3), we propose to investigate novel deep learning based super resolution (SR) image reconstruction networks. The work will include a systematic literature review and the implementation of the most promising SR networks. We have a large clinical dataset of CMRA/CTCA data that can be used for training, testing and validation. It would be useful to know about the fundamentals of artificial intelligence.

Acquisition, reconstruction and motion correction for 3D whole-heart Magnetic Resonance Imaging

Magnetic Resonance Imaging (MRI) is an important non-invasive tool for risk assessment and treatment monitoring of cardiovascular disease. In comparison to other imaging modalities (ultrasound, x-ray and computed tomography) MRI has the advantage of combining excellent soft tissue contrast with high spatial resolution. However, a major limitation of current MRI protocols is that all imaging sequences are acquired sequentially, with different resolution, geometric orientations and breath-hold positions, resulting in long planning and scan times. Moreover, image quality degradation due to respiratory and cardiac motion are major challenges that greatly affect the accuracy and reproducibility of cardiac MRI. This research includes the development of novel whole-heart 3D MRI sequences, motion compensation techniques and deep learning methods to enable 3D, easy-to-plan, fast and efficient quantitative multi-parametric MRI of heart and vessels. The research is done together with King’s College London. Basic knowledge of MRI, signal processing and artificial intelligence would be useful.