Deep Learning Reconstruction for MRI Diffusion Tensor Imaging
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Diffusion tensor imaging (DTI) and fiber tracking methods in magnetic resonance imaging require extensive data acquisition time, limiting their clinical application due to high costs and inefficiencies.
Innovation Solution
A deep learning network is trained using omnidirectionally sampled diffusion weighted images of training samples, allowing for undersampled diffusion weighted images to be used for predicting diffusion tensor images, reducing the need for extensive scanning and improving imaging efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If omnidirectional sampling is used to ensure accurate fiber tracking, then measurement precision is improved, but data acquisition time increases
Solution Approach 1:
A deep learning model is pre-trained using omnidirectionally sampled diffusion weighted images to learn the mapping between undersampled and omnidirectional data. During clinical application, the pre-trained model rapidly reconstructs diffusion tensor images from undersampled data, eliminating the need for time-consuming omnidirectional scanning while maintaining tracking accuracy.
Solution Approach 2:
The patent uses undersampled diffusion weighted images as a simplified copy or representation of the full omnidirectional data. The deep learning model learns to map this reduced data copy to the equivalent omnidirectional diffusion tensor information, enabling accurate fiber tracking without acquiring the complete omnidirectional dataset.
2Reliability
If extensive data acquisition is performed to guarantee fiber tracking accuracy, then reliability is improved, but productivity decreases
Solution Approach 1:
The patent replaces the mechanical data acquisition process (extensive scanning) with an information processing approach (deep learning reconstruction). Instead of mechanically acquiring more data through prolonged scanning, the system uses a trained neural network to computationally generate reliable diffusion tensor images from limited undersampled data, significantly improving imaging efficiency while maintaining reliability.
3Manufacturing precision
If deep learning network is trained with omnidirectional data, then imaging quality is maintained, but device complexity increases
Solution Approach 1:
The deep learning model acts as an intermediary component between the magnetic resonance scanner and the diffusion tensor imaging reconstruction process. This intermediary learns the complex mapping relationships during training and handles the computational complexity internally, allowing the overall system to maintain high imaging quality while presenting a simpler operational interface that only requires undersampled data acquisition.
Data Source
AI summary
A magnetic resonance diffusion tensor imaging method and corresponding device. The method includes acquiring omnidirectionally sampled diffusion weighted images of a plurality of training samples; performing diffusion tensor model fitting and undersampling for the omnidirectionally sampled diffusion weighted images of each training sample to obtain an omnidirectionally sampled diffusion tensor image and an undersampled diffusion weighted image; training a deep learning network, with the omnidirectionally sampled diffusion tensor images of the plurality of training samples as training targets and the undersampled diffusion weighted images as training data; acquiring undersampled diffusion weighted images of a target object; and inputting the undersampled diffusion weighted images of target objects into the trained deep learning network to obtain the predicted omnidirectionally sampled diffusion tensor images of the target objects. Also, a fiber tracking method and corresponding device.