Unsupervised Separation of (Multiple) Pilot Tone Signals From Images Using Weighted PCA

Motivation: Pilot Tone (PT) and Beat Pilot Tone (BPT) signals are used for motion tracking but can cause image artifacts. Efficient separation is key for artifact-free imaging and accurate motion sensing. Goal(s): We develop a method to effectively separate PT and BPT signals from MRI data using weighted Principal Component Analysis (PCA). Approach: We model k-space as a combination of MRI data, pilot tones, and noise. Weighted PCA isolates the motion signal and reconstructs a PT-free image. Results: Weighted PCA successfully separates PT and BPT signals across MRI sequences and experimental conditions, including low-power and multi-tone scenarios, outperforming existing methods in removing PT artifacts. Impact: We present a simple unsupervised approach to separate Pilot Tone signals from MRI data, offering a more accurate representation of PT than Fourier-based methods. It is general and works across scan settings to remove PT leakage artifacts and enable motion-sensing.