Angiographic systems performance evaluation through a channelized hotelling observer model

Medical image quality assessment is a challenging subject that depends on measurable and objective physical properties and long time consuming conditions related to the external perception and human subjectivity. To overcome this issue and to include the human observer in the images validation, observer models are typically used. In this thesis work the Channelized Hotelling Observer (CHO) model is used to test three human observer’s ability in terms of performance in signal detection task on large set of fluoroscopic and angiographic images. Three different systems are used to acquire the images: Siemens Artis Q, Philips Azurion 7 C20 and GE Discovery IGS 740. The last two are angiographic machines available in Reggio Emilia’s hospital used to acquire part of the images of the project. The remaining ones come from an EFOMP’s pre-existing dataset. So, further series of acquisitions are performed on two distinct test objects: TO12 and TO16, that are phantoms appositely realized for contrast-detail (CD) analysis. From CHO application, an α internal noise parameter is estimated and it is used to graph CD curves that are robust tools for evaluating displayed image quality. After a statistical validation of the chosen CD curves, the optimal α are given: in Siemens with TO12, α=3 for fluoroscopy and α=1 for angiography are chosen, in Philips and GE with TO16, phantom limits are achieved in fluoroscopy, in angiography Philips selects α=20 and GE α=10. Comparisons between acquisitions using their optimal CHO CD curves are made and consistent considerations about their statistical significance are done. Indeed, a figure of merit, depending on the contribution of the Kerma Area Product and image quality given by the CD curve, is calculated. The most performant acquisition on Siemens is that with low dose level, FOV=22cm , SID=110cm and pre-filter of 0.6 mm. The best one in angiography between Philips and GE is the low dose level, FOV=32cm, SID=119cm GE acquisition.