Image quality transfer via random forest regression: applications in diffusion MRI.

Area of Science:

• Medical Imaging
• Neuroimaging
• Machine Learning

Background:

• High-quality medical imaging often requires long acquisition times or specialized hardware.
• Enhancing lower-quality standard acquisition data is crucial for broader accessibility and analysis.
• Diffusion MRI data quality can be a limiting factor in detailed structural analysis.

Purpose of the Study:

• To introduce and demonstrate a novel image quality transfer framework for medical imaging.
• To enhance low-quality diffusion MRI datasets using information from high-quality reference datasets.
• To improve super-resolution of diffusion tensor images (DTIs) and enable neurite orientation density and dispersion imaging (NODDI) parameter mapping from standard data.

Main Methods:

• A framework using random forest regression was developed to transfer image quality.
• The method relates image patches from low-quality datasets to voxel values in high-quality datasets.
• The Human Connectome Project (HCP) dataset served as the high-quality reference data.

Main Results:

• The framework successfully enhanced spatial resolution for DTIs in standard datasets.
• Neurite orientation density and dispersion imaging (NODDI) parameter maps were constructed from single-shell HARDI data.
• Experimental results quantified improvements compared to alternative reconstructions and demonstrated efficacy.

Conclusions:

• Image quality transfer is a viable method for enhancing medical imaging data.
• The proposed random forest regression framework effectively improves diffusion MRI data quality.
• This approach broadens the utility of standard MRI acquisitions for detailed neuroimaging analysis.