Data-driven imaging geometric recovery of ultrahigh resolution robotic micro-CT for in-vivo and other applications

TL;DR

This paper presents a novel data-driven approach for geometric correction in ultrahigh-resolution robotic micro-CT imaging, significantly improving image clarity and detail in in-vivo and phantom scans.

Contribution

It introduces a new geometry estimation method using machine learning and normalized cross correlation to reduce artifacts in high-resolution robotic micro-CT imaging.

Findings

Sharper images with finer details after correction

Validated on mouse and phantom scans

Effective in reducing geometric artifacts

Abstract

We introduce an ultrahigh-resolution (50\mu m\) robotic micro-CT design for localized imaging of carotid plaques using robotic arms, cutting-edge detector, and machine learning technologies. To combat geometric error-induced artifacts in interior CT scans, we propose a data-driven geometry estimation method that maximizes the consistency between projection data and the reprojection counterparts of a reconstructed volume. Particularly, we use a normalized cross correlation metric to overcome the projection truncation effect. Our approach is validated on a robotic CT scan of a sacrificed mouse and a micro-CT phantom scan, both producing sharper images with finer details than that prior correction.