GORDA: Graph-based ORientation Distribution Analysis of SLI scatterometry Patterns of Nerve Fibres
This work addresses a limitation in brain imaging for neuroscientists by enabling 3D orientation analysis, though it appears incremental as it builds on existing SLI techniques with a new computational method.
The paper tackled the problem of estimating 3D orientation of nerve fibres from Scattered Light Imaging data, which traditional peak-picking methods could not achieve, by proposing an unsupervised learning approach using spherical convolutions to enable more detailed interpretation of fibre orientation distributions in the brain.
Scattered Light Imaging (SLI) is a novel approach for microscopically revealing the fibre architecture of unstained brain sections. The measurements are obtained by illuminating brain sections from different angles and measuring the transmitted (scattered) light under normal incidence. The evaluation of scattering profiles commonly relies on a peak picking technique and feature extraction from the peaks, which allows quantitative determination of parallel and crossing in-plane nerve fibre directions for each image pixel. However, the estimation of the 3D orientation of the fibres cannot be assessed with the traditional methodology. We propose an unsupervised learning approach using spherical convolutions for estimating the 3D orientation of neural fibres, resulting in a more detailed interpretation of the fibre orientation distributions in the brain.