Docket #: S17-364
MRI Acquisition Trajectory Optimization based on prior knowledge of image content, sensitivity and k-space statistics
Researchers at Stanford developed a method to improve both the efficiency and the performance of MRI optimal trajectory design. Given a reference k-space information, the proposed method uses stochastic optimization to search through possible undersampling trajectory candidates and uses an improved metric to have a quick estimation of the potential performance if each candidate trajectory is applied with PI+CS reconstruction. Our proposed method shows superior performance and better efficiency. It is more than 100x faster than conventional global trajectory optimization methods (Bayesian based method) and is also faster than previously proposed Greedy solution. Compared against related publication on this topic, current method is around 5~10x faster and has around ~20% performance improvement by using better initialization and better optimization methods.
Applications
- Improve acqusition for MRI and similar tomographic acquisition in medical imaging
Advantages
- Our proposed method shows superior performance and better efficiency. It is more than 100x faster than conventional global trajectory optimization methods (Bayesian based method) and is also faster than previously proposed Greedy solution. Compared against related publication on this topic, current method is around 5~10x faster and has around ~20% performance improvement by using better initialization and better optimization methods.
Patents
- Published Application: 20190250233
- Issued: 10,775,466 (USA)
Similar Technologies
-
Improved MRI reconstruction using deep learning, generative adversarial network and acquisition signal model S17-392Improved MRI reconstruction using deep learning, generative adversarial network and acquisition signal model
-
Brain function and connectivity norms - software S17-255Brain function and connectivity norms - software
-
Improved synthetic image contrast generation using Deep Learning S17-454Improved synthetic image contrast generation using Deep Learning