MRI Reconstruction Using Machine Learning Trajectory Optimization

The patent applies parameter changes by optimizing k-space sampling trajectories using machine learning to find optimal sampling patterns that minimize artifacts. The system learns to adjust trajectory parameters such as sampling density, acceleration factors, and k-space coverage to improve image quality while reducing harmful artifacts.

The patent replaces traditional mechanical/optical reconstruction methods with machine learning-based reconstruction algorithms. The system uses trained neural networks to reconstruct images from undersampled k-space data, substituting conventional iterative reconstruction methods with learned models that reduce computational burden and artifact production.

The patent applies preliminary action by pre-training machine learning models on large datasets of k-space trajectories and corresponding images before actual MRI reconstruction. The system performs computational work in advance to learn optimal reconstruction patterns, so that during actual MRI scanning, reconstruction can be performed rapidly using the pre-trained models rather than computing from scratch.

The patent uses copying by creating trained machine learning models that capture reconstruction patterns from training data. Once trained on comprehensive datasets, these model copies can rapidly reconstruct images from new k-space data without requiring the full computational resources of the original training process, enabling fast reconstruction with reduced processing power.