[OA096] Low contrast detectability assessment of dynamic angiographic images: spatio-temporal generalisation of model observer and comparison with 2AFC experiments

Purpose Evaluate the feasibility of spatio-temporal generalisation of mathematical approaches for protocol optimisation in interventional radiology. Methods A Allura FD20 (Philips, The Netherlands) has been employed in this study. Two different Model Observers has been considered: Non-PreWhitening Eye filter (NPWE) with spatio-temporal contrast sensitivity function and Channelized Hotelling Observer (CHO) with spatio-temporal Gabor filters. Furthermore, Low Contrast Detectability has been evaluated with a statistical method using uniform images. This analysis has been performed on an averaging of frames depending from the visualised fps. In particular, the averaging process was performed on groups of frames that correspond to the noise integration time of human eye-brain system of 200 ms, e.g. 3 frames with 15 fps. A series of 2AFC experiments have been performed by two observers. The evaluation and characterization of the mathematical approaches were performed throughout two different phases: firstly, a series of simulated images of the insert were used to characterize the models. A total of 3 insert dimensions (1.25, 2, 3 mm of diameter), 25 contrast levels and 4 different visualised frame rate (5, 10, 15 and 30 fps) have been selected to characterise low contrast detectability. Results of this analysis were than used to evaluate low contrast detectability with the acquired images of a dedicated commercial phantom CDRAD phantom (Artinis Medical Systems, The Netherlands). Four modalities were selected in this study. A total of 3 insert dimensions (1.3, 2, 3.2 mm of diameter), 15 contrast levels have been selected for every considered modality. Results Analyses with simulated images underline a good agreement ( Conclusions The agreement with human observer experiments underlines the feasibility of the proposed generalisations. Thus, they could be introduced for a deeper modalities characterisation, in terms of image quality, or for clinical protocol optimization involving cine images.