Skip to main content
Photon Sciences

Calibrating Tomography Data Using Machine Learning

An Argonne team has developed a machine learning approach for calibrating the center of rotation in x-ray light source tomography data that provides better accuracy than conventional imaging processing-based methods.

A convolutional neural network (CNN) is a robust approach to determning rotation axis in tomographic imaging. Argonne has demonstrated that a CNN classification model to calibrate the center of rotation works accurately, once it has been trained with sufficient prior estimation. Evaluation with synthetic phantoms shows good accuracy for different noise conditions. The approach has been validated on x-ray synchrotron light source data and provides better accuracy than image registration-based methods and entropy-based methods.

This machine learning approach also has great potential for reducing or removing other artifacts caused by instrument instability, detector non-linearity, etc. An open-source toolbox that integrates CNN methods is freely available through GitHub at tomography/xlearn. The toolbox can be easily integrated into existing computational pipelines available at synchrotron light source facilities.