Robust Subspace Clustering Approach for High-Dimensional MRF: Novel Simultaneous Clustering and Dimensionality Reduction at Scale

Motivation: Magnetic Resonance Fingerprinting (MRF) has become a promising technique for simultaneously estimating several MR tissue parameter maps. However, as the number of parameters increases, existing models encounter significant challenges related to computational power and storage. Goal(s): To address the challenges associated with high-dimensional MRF data by developing a clustered low-rank representation of MRF signals. Approach: We introduce a novel method for robust, simultaneous clustering and dimensionality reduction of high-dimensional MRF signals. This approach leverages Student's t-mixtures, to enhance performance. Results: Our method significantly reduces the time complexity of traditional matching processes for reconstructing maps across six parameters, advancing MRF closer to clinical applicability. Impact: This work tackles the computational challenges inherent to MRF for various tissue parameters, including relaxometry, magnetic field characteristics, and microvascular properties. While the method is demonstrated on MRF reconstruction, it can be applied to other large-scale dimensionality reduction tasks.