Image quality transfer and applications in diffusion MRI.

Area of Science:

• Medical Imaging
• Computational Imaging
• Machine Learning

Background:

• Routine medical imaging acquisitions often yield lower-quality data compared to specialized experimental devices.
• There is a need to leverage the rich information from high-quality imaging for routine data analysis.

Purpose of the Study:

• To introduce and demonstrate a novel computational imaging technique, image quality transfer (IQT).
• To utilize machine learning for transferring image quality from high-quality to low-quality medical imaging data.
• To showcase IQT's application in enhancing diffusion MRI data for brain connectivity and microstructure imaging.

Main Methods:

• Developed a machine learning approach (IQT) using matched pairs of low-quality and high-quality images to learn data mappings.
• Implemented a patch-regression model for IQT.
• Applied IQT to diffusion MRI datasets from the Human Connectome Project (HCP).

Main Results:

• IQT successfully enhanced image resolution and information content in unseen low-quality images.
• Demonstrated improved brain connectivity mapping by revealing thin pathways from standard datasets.
• Showcased potential in estimating microstructural parameters from single-shell diffusion MRI data, usually requiring multi-shell data.
• Exhibited strong generalizability of IQT even with diverse training datasets.

Conclusions:

• Image quality transfer (IQT) is a versatile machine learning technique for enhancing medical imaging data quality.
• IQT offers significant potential benefits for brain connectivity mapping and microstructure imaging using diffusion MRI.
• The IQT concept is broadly applicable to various imaging modalities and reconstruction challenges.