ANN-Based MRI Pulse Sequence Design for MRF Signal Evolution
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional manual approaches for designing magnetic resonance imaging (MRI) pulse sequences are complex and time-consuming, limiting their ability to produce signal evolutions with desired characteristics, particularly for magnetic resonance fingerprinting applications, which require varied sequence blocks with multiple parameters.
Innovation Solution
An artificial neural network (ANN) is trained using transverse magnetization signal evolutions with arbitrary initial magnetizations to automatically design MRI pulse sequences with user-controllable settings such as acquisition period and flip angle, enabling the production of a wider array of signal evolutions suitable for magnetic resonance fingerprinting.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If conventional manual approaches are used to design MRI pulse sequences, then the design process can be performed with simple constant parameters, but the process becomes complex and time-consuming, limiting productivity
Solution Approach 1:
The patent replaces manual mechanical design processes with an artificial neural network system. The ANN automatically generates pulse sequences by learning from training data, substituting the manual iterative design process with an automated computational approach that rapidly produces sequences with desired signal characteristics
Solution Approach 2:
The invention transforms the design approach from using constant parameters to using varied parameters. The ANN-generated sequences incorporate varying acquisition periods and flip angles, enabling the production of diverse signal evolutions suitable for magnetic resonance fingerprinting while maintaining design efficiency
2Adaptability or versatility
If conventional ANN approaches are used with known pulse sequences for training, then the training process is constrained to limited signal evolutions, but the approach is simpler to implement
Solution Approach 1:
The patent applies preliminary action by pre-generating a comprehensive library of signal evolutions with arbitrary initial magnetizations before training the ANN. This pre-computed training data enables the network to learn a broader range of signal behaviors, making the system adaptable to various MRF applications without constraining the signal evolution range
Solution Approach 2:
The invention creates a universal training approach where the ANN is trained on diverse signal evolutions that can accommodate multiple initial magnetization conditions. This universal training data structure allows the single ANN model to generate sequences suitable for various MRF applications, enhancing versatility without requiring multiple specialized models
Data Source
AI summary
Example apparatus and methods employ an artificial neural network (ANN) to automatically design magnetic resonance (MR) pulse sequences. The ANN is trained using transverse magnetization signal evolutions having arbitrary initial magnetizations. The trained up ANN may then produce an array of signal evolutions associated with a pulse sequence having user selectable pulse sequence parameters that vary in degrees of freedom associated with magnetic resonance fingerprinting (MRF). Efficient and accurate approaches are provided for predicting user controllable MR pulse sequence settings including, but not limited to, acquisition period and flip angle (FA). The acquisition period and FA may be different in different sequence blocks in the pulse sequence produced by the ANN. Predicting user controllable MR pulse sequence settings for both conventional MR and MRF facilitates achieving desired signal characteristics from a signal evolution produced in response to an automatically generated pulse sequence.