AFT-Net MRI Reconstruction for Noisy K-Space Mapping
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional MRI image reconstruction methods struggle with noise and distortion due to random error and lack generality, especially when dealing with raw data from MRI scanners, leading to unreliable and inconsistent image quality.
Innovation Solution
A unified complex-valued image reconstruction approach using the Artificial Fourier Transform (AFT) framework, integrated with deep learning networks like AFT-Net, to learn the mapping between k-space and image domains, effectively removing noise and preserving structural information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional Fourier transformation methods are used for MRI reconstruction, then the reconstruction process is straightforward and computationally efficient, but the image quality suffers from noise and distortion due to random error
Solution Approach 1:
The patent introduces an Artificial Fourier Transform (AFT) network as an intermediary between raw k-space data and final image reconstruction. This neural network mediator learns to map k-space data to image domain while inherently handling noise and distortion, resolving the contradiction by placing a smart intermediary layer that improves reliability without requiring complete redesign of the reconstruction pipeline
Solution Approach 2:
The AFT network employs fully adjustable parameters that can be fine-tuned through training on specific datasets. By changing the parameters of the transformation function from fixed mathematical constants to learnable parameters, the system adapts to reduce noise and distortion while maintaining reconstruction accuracy, thus improving image quality without excessive complexity
2Loss of information
If domain conversion from k-space to image domain is performed using conventional methods, then the process is simple and fast, but information loss occurs inevitably during the transformation
Solution Approach 1:
The AFT network performs preliminary learning of the k-space to image domain mapping during a training phase. By pre-learning the optimal transformation that preserves information, the network embeds this knowledge into its parameters, allowing fast inference without information loss during actual reconstruction, thus resolving the time-quality tradeoff
3Reliability
If deep learning networks with multiple layers are used for MRI reconstruction, then noise removal and structural preservation improve, but the computational complexity and training requirements increase significantly
Solution Approach 1:
The reconstruction task is segmented into distinct functional modules within the AFT network: k-space data input, complex-valued feature extraction layers, transformation blocks, and image domain output. This segmentation allows each module to be optimized independently for noise removal while controlling overall computational energy consumption through selective training and inference
Data Source
AI summary
Disclosed are methods, systems, and other implementations, including a unified complex-valued deep learning framework (AFT-Net), which determines the k-space domain to image domain mapping for MRI reconstruction and allows incorporation of existing deep learning models. Embodiments include a computer-implemented method for reconstructing images that includes obtaining resonance (MR) k-space data resulting from a scan performed by an MRI scanner on tissue of a patient, with the MR k-space data including complex-valued data, and processing, by a complex-valued machine learning image reconstruction system, the complex-valued data of the MR k-space data to generate image data representing features of the MR k-space data. The processing may include performing data filtering operations, by one or more machine learning filter blocks implemented according to a CU-Net architecture realized using one or more convolutional neural networks (CNN) configured for complex data processing, on data that is based on the k-space data.