MR Learned Pulse Sequence on Bi-Exponential 3D T2 and T1rho Mapping of Healthy and Early Knee Osteoarthritis: Linear Discriminant Analysis

Motivation: Bi-exponential (BE) T2 and T1ρ mapping show promise for early knee osteoarthritis (OA) detection, but long scan times and low SNR limit their usage. Goal(s): Use our learned pulse sequence (L-MPGRE) for T2 and T1ρ mapping and evaluate the discrimination between OA patients and healthy subjects (HS). Approach: We used the L-MPGRE sequence on 5 OA patients (KL=1 or 2) and 6 HS to measure T2 and T1ρ values, and compute effect-size and p-values. Results: T1ρ value is better at discriminating early OA from HS. However, successful differentiation with BE models depends on using linear discriminant analysis and proper threshold of BE components. Impact: Learned pulse sequences are useful for early OA detection, with T1ρ outperforming T2 mapping. Results indicate that BE models can be advantageous if linear discriminant analysis and the proper threshold are used.