Crystalline Neural Network: a diffusion weighted neuroimage processing approach
Abstract
Diffusion weighted imaging is a magnetic resonance technique especially sensitive to the water molecular diffusion. By fitting biophysical models to the acquired data is possible to describe the regional direction of the water movement. This technique is widely used to study the nervous tissue microstructure and brain connectivity. We present a new diffusion model, inspired by artificial neural networks, but using a crystalline architecture. The model is presented in addition with a fitting method. To evaluate its performance, we performed several tests on virtual phantoms which mimic probable nerve fiber layouts.
The model performed well on phantoms data and showed potential to resolve fiber dispositions better that other standard diffusion models, such as crosses and curves in fibers path.