PET/MR Attenuation Correction Using R1-Guided DL-TESLA Maps
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current MRI-based attenuation correction methods for PET/MR imaging lack direct association with electron density, leading to inter-subject and across-center variations, and there is a need for improved test-retest repeatability to distinguish true pathophysiological changes from measurement error.
Innovation Solution
A deep learning-based method, DL-TESLA, uses MR images of longitudinal relaxation rates (R1) and a 3D residual UNet model to transform pseudo-CT maps into linear attenuation coefficients, balancing representations of bone, air, and soft tissue regions, enhancing accuracy and repeatability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional MRI-based AC methods (atlas, direct imaging/segmentation) are used, then the process is simpler and faster, but accuracy and precision are insufficient leading to inter-subject and across-center variations
Solution Approach 1:
The patent transforms qualitative MR signal intensities into quantitative R1 relaxation rates, which have a direct linear relationship with electron density and CT Hounsfield units. This parameter transformation enables accurate LAC derivation while maintaining methodological simplicity through the established R1-CT correlation
Solution Approach 2:
The patent introduces R1 relaxation rate maps as an intermediary between MR imaging and attenuation correction. These R1 maps serve as a quantitative bridge that directly correlates with electron density and CT values, enabling accurate LAC generation without complex atlas matching or segmentation procedures
2Measurement precision
If deep learning methods are used to improve accuracy, then measurement precision improves, but processing time and computational complexity increase
Solution Approach 1:
The patent extracts and utilizes only the essential R1 relaxation rate parameter from MR imaging for attenuation correction, avoiding the need for complex deep learning architectures. This extraction of the key quantitative parameter achieves high accuracy while maintaining rapid processing speeds
Solution Approach 2:
By transforming MR data into R1 relaxation rates and utilizing the linear R1-CT relationship, the patent achieves accurate LAC derivation through straightforward parameter transformation rather than computationally intensive deep learning models, thus reducing processing time while maintaining precision
3Measurement precision
If quantitative MRI relaxation rates (R1) are used instead of arbitrary MR signal intensities, then measurement precision and repeatability improve, but the complexity of data processing increases
Solution Approach 1:
The patent transforms arbitrary MR signal intensities into quantitative R1 relaxation rates through standardized processing. The resulting R1 maps have a direct linear relationship with electron density and CT values, improving repeatability while the processing complexity is managed through established quantitative MRI protocols and the simple R1-CT correlation model
Data Source
AI summary
Systems and methods for deep-learning-based T1-enhanced selection of linear attenuation coefficients (DL-TESLA) for PET/MR attenuation are described.