OSCAR: Online Spectrum Classification and Review
One of the major hurdles in the analysis of magnetic resonance spectroscopic images is being able to assess whether or not a spectrum is of good quality. A skilled clinician is able to do this very well for a given spectrum – but there can be hundreds of thousands of spectra in a high-resolution image! Therefore, we are working on developing an algorithm which can automatically filter out poor quality spectra. To do this, we are using deep learning to build a model that can attempt to mimic how a radiologist may look at a particular spectrum. The first step in this endeavor is collecting training data.
The OSCAR platform is a web-based tool designed by the Brain Imaging Group to generate training data sets for the deep learning algorithm. Our goals are for the spectrum analyzer to:
- randomize thousands of spectra from various types of brain tissue (normal and pathologic) and from different subjects
- distribute these spectra to skilled radiologists from across the country to assess, ensuring that any single spectrum is viewed by multiple clinicians
- coalesce the data and build consensus models
- be easily distributable for other institutions to use
We are currently developing OSCAR and testing it in-house at Emory. If you are interested in participating in a beta test of the tool, please contact Saumya Gurbani.
If you already have credentials to log in to OSCAR, please click on the link below to begin classification: