Model observers for Low Contrast Detectability evaluation in dynamic angiography: A feasible approach

Abstract

PurposeTo evaluate the feasibility of spatio-temporal generalisation of mathematical methods for protocol optimisation in interventional radiology. Materials and methodsTwo model observers were considered:•Channelized Hotelling Observer with spatio-temporal Gabor channels;•Non-Prewhitening Model Observer with spatio-temporal contrast sensitivity function.Furthermore, Low Contrast Detectability (LCD) was evaluated with a generalised statistical method by taking into account the noise integration capability of the human eye.A series of two alternative force choices (2AFC) experiments performed by four observers were used to evaluate the reliability of the proposed models.The evaluation of the mathematical methods was performed by comparing their results to the human observer performances in two steps:1. Firstly, a series of simulated images were used to tune the models2. In the second phase, tuned models were applied both to simulated images and actual images obtained with a commercial phantom to evaluate detectability scores. ResultsEvaluation with simulated images shows a good agreement with 2AFC results (RMSE < 10%). Phantom-based evaluations show a general decrease of such agreement, characterized by an RMSE lower than 16%. ConclusionsThe agreement with human observer experiments supports the feasibility of the proposed generalisations. Thus, they could be introduced in quality control programmes for a deeper protocol-characterisation or for clinical protocol-optimization when dynamic images are involved.