Spiral-PRIME

Spiral-PRIME: Physics-based Reconstruction with Iterative Model-based Enhancement

Spiral-PRIME is a hybrid 2D/3D reconstruction framework designed for localized quadratic (LQ) RF encoded spin-echo imaging. It is implemented within the Graphical Programming Interface (GPI) environment to provide high-resolution volumetric brain MRI using spiral trajectories as a clinically viable alternative to 3D Fast Spin-Echo (FSE).

Key Features

• Hybrid 2D/3D Reconstruction: Joint spiral deblurring, fat-water separation, and 3D wavelet-based noise suppression within a unified ADMM optimization.

Reconstruction Architecture

The Spiral-PRIME reconstruction is built on the Graphical Programming Interface (GPI) framework. It solves the following inverse problem using the Alternating Direction Method of Multipliers (ADMM):

$$ \min_{x} \frac{1}{2}|\tilde{E}x-d|_{2}^{2} + \sum_{c\in{w,f}} \lambda_{c} |\Psi x_{c}|_{1} $$

Where $\tilde{E}$ is the forward operator, $x$ contains the water and fat images, and $\Psi$ represents the 3D wavelet transform.

Figure 3: Schematic of the Spiral-PRIME reconstruction framework.

• (a) Preprocessing includes noise whitening, coil compression, LQ + sliding-slice correction, and concomitant field correction.
• (b) The iterative ADMM loop performs slice-wise data consistency via conjugate gradient and volumetric 3D wavelet denoising.

System Requirements

• Operating System: macOS is required for compatibility with the GPI implementation used in this study.
• Hardware: Developed and validated on Apple Silicon (M2 Pro).
• Software: GPI Framework must be installed (https://github.com/gpilab/framework).
• Compilers: A C++ compiler is necessary for multi-threaded slice processing.

Installation and Build

Clone the repository (development branch):

git clone -b develop https://github.com/gpilab/spiral-prime.git

Navigate to project folder:

cd spiral-prime

Build the code using the GPI build utility:

gpi_make

(Note: active development is on the develop branch; stable releases will be pushed to main.)

Results

Spiral-PRIME achieves consistent T2w SNR gains of 30–40% compared to linear spiral-SENSE reconstructions while preserving fine structural fidelity at $800,\mu\text{m}$ isotropic resolution.

Figure 5: Accelerated T2w results comparing spiral-SENSE and Spiral-PRIME, highlighting significant noise suppression and artifact mitigation.

Citation

If you use this framework or the provided data, please cite:

Krishnamoorthy G, Velikina JV, Pipe JG. Localized Quadratic RF encoded Spin-Echo with Spiral-PRIME Reconstruction: A Practical Alternative to 3D FSE for High-Resolution Volumetric Brain MRI. Magnetic Resonance in Medicine (2026)

---

For more details about implementation and the GPI-based reconstruction pipeline, see the source files in src/ and the GPI build outputs in the build/ directory.