PROJECT TITLE :

DeepSPIO Super Paramagnetic Iron Oxide Particle Quantification using Deep Learning in Magnetic Resonance Imaging

ABSTRACT:

Because of their susceptibility, super paramagnetic iron oxide (SPIO) particles can be utilized as a useful contrast agent for a variety of purposes within the realm of MRI. Quantification of these particles is typically performed using relaxometry or by measuring the inhomogeneities produced by the particles in question. These methods are dependent on the phase, which produces unreliable results when applied to high concentrations. This study presents a novel Deep Learning method that was developed by us in order to quantify the SPIO concentration distribution. We used a new sequence that we developed called View Line to acquire the data. Within this sequence, the information about the field map is encoded within the geometry of the image. Our network is unique in that it employs residual blocks to act as bottlenecks and multiple decoders to enhance the gradient flow within the network. This is the network's most notable innovation. Each decoder is able to make a unique prediction regarding a particular aspect of the wavelet decomposition of the concentration map. This decomposition not only speeds up the process of the model's convergence, but it also results in a more accurate estimation of the concentration. We put our technique for SPIO concentration reconstruction through its paces by simulating images and using data obtained from actual scans of phantoms. The simulations were carried out with the help of the images from the IXI dataset, and the phantoms were made up of plastic cylinders that contained agar mixed with SPIO particles in varying concentrations. In both of the experiments, the model was successful in providing accurate quantifications of the distribution.