Multi-parametric MRI techniques quantify multiple physical parameter values, such as T₁ and T₂, simultaneously. They do so by computing a dictionary of expected complex signal vectors over a predetermined set of parameter value combinations ahead of time. Parameter maps are then reconstructed by finding for each voxel the most similar signal in the precomputed dictionary and assigning its parameter value combination to that voxel. This approach has been described in the literature as magnetic resonance fingerprinting (MRF).

Most research on MRF is done with clinical scanners at magnetic fields beyond 1 Tesla. Our goal is to adapt these techniques to a low-field regime in order to make use of the increased physical accessibility and reduced cost offered by low-field scanners.

MRF generally employs measurement sequences that follow pseudorandom variations of the scanner control parameters. The actual schedule by which the scanner control parameters are varied impacts the ability of the measurement to discriminate between similar tissue parameters thus impacting the achievable resolution of tissue parameter values.

At the AMT Center, we develop methods to optimize MRI sequence parameters towards maximizing the discrimination of similar tissue parameters.

The signal similarity search required for reconstructing multi-parametric MRI data is computationally expensive and limits the reconstructable number of parameters and their resolution.

Here, we leverage approaches based on similarity search originally developed for searching vast multimedia collections to accelerate the reconstruction.