Unsupervised Reconstruction for Ungated Ghost Angiography by Clustering of Image Features

Introduction: Ghost magnetic resonance angiography (MRA) has been proposed as an unenhanced and ungated method for angiography that is based on the presence of ghost artifacts resulting from pulsatile blood flow (1). Although the method is simple to acquire in that cardiac gating is not required, it requires manual post-processing to identify suitable slices in a large stack from which to create an interpretable angiogram. To maximize the contrast of the final angiogram it is necessary to eliminate slices located within the body and to carefully select the slices that contain conspicuous ghost artifacts. This manual post-processing step is time-consuming and can introduce unwanted interand intraobserver variability. The purpose of this work was to completely automate the reconstruction process during ungated and non-contrast-enhanced Ghost MRA using image analysis and clustering.