Abstract
Magnetic resonance (MR) tagging is an imaging technique for noninvasively tracking tissue motion in vivo by creating a visible pattern of magnetization saturation (tags) that deforms with the tissue. Due to longitudinal relaxation and progression to steady-state, the tags and tissue brightnesses change over time, which makes tracking with optical flow methods error-prone. Although Fourier methods can alleviate these problems, they are also sensitive to brightness changes as well as spectral spreading due to motion. To address these problems, we introduce the brightness-invariant tracking estimation (BRITE) technique for tagged MRI. BRITE disentangles the anatomy from the tag pattern in the observed tagged image sequence and simultaneously estimates the Lagrangian motion. The inherent ill-posedness of this problem is addressed by leveraging the expressive power of denoising diffusion probabilistic models to represent the probabilistic distribution of the underlying anatomy and the flexibility of physics-informed neural networks to estimate biologically-plausible motion. A set of tagged MR images of a gel phantom was acquired with various tag periods and imaging flip angles to demonstrate the impact of brightness variations and to validate our method. The results show that BRITE achieves more accurate motion and strain estimates as compared to other state of the art methods, while also being resistant to tag fading.
The Challenge: Tag Fading and Spectral Overlap
Magnetic Resonance (MR) tagging is a vital tool for non-invasively "painting" a grid or pattern onto living tissue to observe deformation. However, automated tracking faces two major physical hurdles. First, tag fading occurs due to T1 relaxation, where high-contrast tags fade into the background over the imaging cycle. This violates the "brightness constancy" assumption required by traditional optical flow methods. Second, spectral overlap complicates Fourier-domain methods. As tissue moves and tags fade, the harmonic peaks spread and overlap with the central anatomical signal, making it difficult to isolate motion.
The BRITE Framework
To overcome these issues, the Brightness-Invariant Tracking Estimation (BRITE) framework approaches motion tracking as a disentanglement task. It consists of three primary components:
- Anatomical Disentanglement: The model separates the initial image ($t=0$) into an "anatomy" component and a "tag pattern." A Denoising Diffusion Probabilistic Model (DDPM) serves as a powerful prior to ensure the reconstructed anatomy is realistic, while a CNN encoder predicts tag parameters (frequency, phase, amplitude).
- Lagrangian Motion Tracking: A Physics-Informed Neural Network (PINN) estimates a stationary velocity field. This field is integrated to produce a diffeomorphic mapping, ensuring the estimated motion is smooth and biologically plausible.
- Tag Fading Module: This component learns the evolution of tag amplitude and DC offset at every time step, allowing the model to reconstruct faded images based on estimated motion.
BRITE is entirely unsupervised, optimizing its parameters directly on the test sequence without requiring ground truth labels.
Experimental Validation
The framework was evaluated using a physical silicone gel phantom scanned with varying tag periods (9mm to 26mm) and flip angles. These variations simulated different levels of signal degradation and tag fading over a 1.1-second sequence.
Key Results and Insights
BRITE demonstrated superior accuracy compared to state-of-the-art methods like HARP, SinMod, and DeepTag in both End-Point Error (EPE) and Maximum Principal Strain (MPS) metrics. Key findings include:
- Robustness to Fading: Unlike Fourier-based methods where errors spike as tags fade, BRITE maintains low and stable error rates throughout the sequence.
- Handling Large Tag Periods: BRITE excels in scenarios with large tag periods (e.g., 26mm) where spectral overlap is most severe and traditional filters often fail.
- Implicit Smoothness: The PINN architecture provides inherent regularization, avoiding the need for manual smoothness penalties common in traditional registration.
While currently validated in 2D on gel phantoms, the authors suggest that extending BRITE to 3D and clinical applications, such as cardiac or tongue tracking, is a promising direction for future research.
Summary updated