CT perfusion studies of lung cancer: automatic detection of misleading structures and artefacts

Purpose: The aim of this work is to detect and highlight blood vessels, artefacts, and statistically unreliable blood flow values in CT perfusion (CTp) studies of lung cancer through automatic analysis of the Time-Concentration Curves (TCCs). . Methods and Materials: 16 patients with primary lung tumour underwent axial CTp, for a total amount of 24 examinations. Blood flow values were computed on fitted data after motion correction, according to the maximum slope method. The average error of the fitted TCC model with respect to the original Hounsfield Unit (HU) values are computed for each voxel and gathered into a histogram. An adaptive parametric threshold was conceived, allowing the automatic selection of voxels in perfusion maps whose model fit error is above the threshold. This study was approved by the institutional review board. Results: Most of the highlighted voxels appeared to be arranged into connected regions, the nature of which is confirmed by two 25-year experienced radiologists operating in a blinded fashion. In particular, these regions resulted to be either physical structures, such as bronchi or vessels, or artefacts coming from reconstruction or residual motion. Conclusions: The presence of vessels, bronchi or artefacts in perfusion maps alters the right perception of the perfusion pattern by radiologists, besides jeopardizing results from any subsequent computation or statistical analysis. In addition, the automatic exclusion of these misleading values prevents radiologists from misinterpreting the perfusion maps, possibly leading to wrong clinical considerations, this representing a step forward to clinical utilization of CTp.