2962
View Abstract / View Presentation
Improving Fast 3D-T1rho Mapping of Human Knee Cartilage with Data-Driven Learned Sampling Pattern
Marcelo Victor Wust Zibetti1, Azadeh Sharafi1, and Ravinder Regatte1
1Radiology, NYU Langone Health, New York, NY, United States
  • The learned sampling pattern (SP) improved the median of the normalized absolute deviation (MNAD) in accelerated T1rho mapping by 7% on average, with a maximum improvement of 21.5% in the testing data. The largest improvement was achieved for 16-fold acceleration.
Figure 5: Representative T1rho maps of the medial knee joint at acceleration factor (AF) 12. The visual difference in the medial cartilage is more noticeable at this AF. The optimized SP is more important at high AFs.
Figure 2: a. Poisson disk sampling pattern (SP), used as an initial SP and b-d. SP obtained from optimization with bias-accelerated subset selection for CS with b. LR, c. L+S SFD, and d. STFD.