Emerging low-field Magnetic Resonance Imaging (MRI) systems offer the promise of low-cost point-of-care imaging that could be conducted in, for example, rural locations, on the battlefield, and eventually even in an ambulance. However, present low-field (LF) MRI results in poor quality images due to low spatial resolution and high noise. These qualities prevent quantitative analyses of images, which are critical for advanced diagnostics.
We are using artificial intelligence (AI) methods to restore image quality and enable quantitative mapping of low-field diffusion MRI images. In collaboration with the Magnetic Imaging Group in PML, and with the Mathematical Analysis and Modeling Group in ITL, we will model low-field brain images from high-field diffusion images, and then design AI networks to transform the lower quality low-field images into Apparent Diffusion Coefficients (ADC) maps to quantify density measurements of water in the brain.
The overall goal of this project is to use an AI network to overcome the low Signal-to-Noise-Ratio (SNR) of low-field MRI, so that we can produce quantitative ADC measurements. This will be accomplished by a collaborative effort imaging with a low-field Hyperfine instrument, with the physicists benchmarking using diffusion phantoms, the mathematicians modeling the physics of differentiating high and low field images, and our group creating the AI models. ADC maps are created from diffusion images taken at a range of magnetic b-values, and our new network design will take in these sets of diffusion images to produce a single ADC map.
This project was initiated 10/2020, and we have been working with different types of image-to-image regression networks that can taken in multiple diffusion images and produce an ADC map using high-field images of the PML’s diffusion phantom. We have shown that the network is capable of the type of exponential approximations required to produce such a map.