Automatic computation of liver and lung perfusion parameters through the analysis of CT image sequences

Abstract: Purpose: The goal of this PhD Thesis work is contributing to advance Computed Tomography perfusion (CTp) towards standardization by improving the accuracy of perfusion results (1), the clinical relevance of studies where global perfusion parameters are commonly utilised (2) and the reproducibility of results in multicentre studies (3). Methods and Materials: 315 liver CTp examinations from 14 Centres and 34 lung CTp studies were considered. First, through a voxel based analysis of the time-concentration curves, an error index capable to measure the quality of perfusion results was set up and validated. Thereafter, an adaptive algorithm exploiting that error index was developed to automatically detect on CTp maps misleading perfusion values (1). Then, a measure of tumour functional heterogeneity was conceived (2). Finally, data pertaining to a large multicentre study on liver CTp (PIXEL) were analysed to also investigate the effects of the sources of protocol heterogeneity (3). Results: The algorithm developed was correctly able to identify and exclude from the analysis vessels, bronchi, and artefacts, allowing to improve perfusion results reliability (1). Global values cannot take into account haemodynamic heterogeneity proving to even mislead clinical considerations (2). Tentative guidelines were provided to help planning protocols (3). Conclusion: Reliability of results improved by detecting and removing unreliable CTp values. A measure of functional heterogeneity must be provided together with mean perfusion values to improve the clinical representativeness of the studies. In conclusion, the automatic methods implemented and the tentative guidelines for multicentre studies represent a clear step forwards to CTp translation in the standard clinic.